US20110208357A1 - Autonomous Mobile Robot - Google Patents

Abstract

A mobile robot is equipped with a range finder and a stereo vision system. The mobile robot is capable of autonomously navigating through urban terrain, generating a map based on data from the range finder and transmitting the map to the operator, as part of several reconnaissance operations selectable by the operator. The mobile robot employs a Hough transform technique to identify linear features in its environment, and then aligns itself with the identified linear features in order to navigate through the urban terrain; while at the same time, a scaled vector field histogram technique is applied to the combination of range finder and stereo vision data to detect and avoid obstacles the mobile robot encounters when navigating autonomously. Also, the missions performed by the mobile robot may include limitation parameters based on distance or time elapsed, to ensure completion of the autonomous operations.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS
• This is a continuation of U.S. patent application Ser. No. 11/618,742, filed Dec. 30, 2006, published as Pre-grant Patent Application Publication 2007/0156286 on Jul. 5, 2007, which is a non-provisional of U.S. Provisional Patent Application No. 60/754,635, filed Dec. 30, 2005, and both applications are incorporated by reference herein.
BACKGROUND
• In the field of mobile robots, mobile robots have been increasingly used in hostile environments (such as, for example, in battle conditions or in rescue operations) for tasks such as ordnance or explosives handling or disposal, field reconnaissance, terrain mapping, and various other procedures in which a considerable risk to life or safety would exist if a person were to perform the task. As a non-limiting example, remotely controlled mobile robots equipped with video cameras have been used to scout hostile positions in environments such as caves or other non-urban terrain. However, limitations of such teleoperated mobile robots include restriction of the operative range of such robots to line-of-sight or areas substantially proximal to the tele-operator, because of real-time wireless communications issues (for example, limited wireless bandwidth availability and/or transmission power for transmitting live camera signals to a teleoperator station) presented by robots that cannot operate autonomously but instead rely entirely on continuous real-time remote control. As a result, risks to the safety of the teleoperator may be heightened because of the necessity to remain rather close to such a remote-controlled mobile robot during operation.
• Furthermore, urban warfare is becoming increasingly important in many military operations. For example, combat in cities tends to generate increased risk factors to military forces, such as high risk of “friendly fire” and non-combatant casualties. However, urban combat may be increasingly common because of political and sociological reasons, and adversaries may make extensive use of guerrilla warfare and other unconventional tactics in heavily developed areas such as cities, industrial areas, and critical infrastructure such as paved roadways or neighborhood streets. At the same time, military forces may be required to take extreme care to minimize civilian casualties and other collateral damage.
• Under these conditions, in order to maintain situational awareness, aerial and satellite reconnaissance can provide valuable high-level information about terrain and troop movements; however, even UAVs and other low-flying aircraft are limited in their ability to provide real-time information to ground troops about what lies around the next corner or on the next block. Soldiers currently perform most dangerous reconnaissance tasks themselves, potentially placing themselves at great risk in hostile urban environments.
• Recently, however, mobile robot platforms are increasingly commonly being deployed by military forces for reconnaissance and other tasks in dangerous environments. As noted above, non-autonomous tele-operated mobile robots have the potential to reduce the risk to military warfighters in urban environments, but they are limited by both radio range, interference and the need for a full-time operator. In urban environments, for example, radio signal attenuation caused by buildings or radio interference may substantially reduce the operational range of such non-autonomous, remote-controlled robots. In addition, the need for a robot operator to devote continuous attention to operating the robot throughout the duration of an operation increases the manpower requirements associated with robotic reconnaissance.
SUMMARY
• In view of the above, man-portable mobile robots (also referred to herein as “unmanned ground vehicles” or “UGVs”) may be enabled to perform fully-autonomous or semi-autonomous reconnaissance missions in urban, industrial or other such developed environments. The mobile robots may be able to explore beyond the range of radio communications required for non-autonomous robots, generate detailed maps of complex urban terrain, record or transmit video and infrared image data, and return to provide up-to-date reconnaissance information for military warfighters.
• Substantial academic research in the field of robotics has previously been conducted on the subjects of mapping and localization using mobile robots. However, the previous research has typically attempted to elucidate complex, monolithic, “kitchen sink”-type general approaches for autonomous robot navigation, while the few robot systems based thereon that have actually been constructed have been significantly limited in their real-world robustness and utility. One reason for this is because the complex algorithms and control processes contemplated by academic researchers often require computational or sensory capabilities that are unrealistic for a man-portable military-use mobile robot system due to expense, weight, and fragility of necessary components.
• Rather than attempting to solve the unbounded, abstract question of robot navigation in the general case, presently discussed are autonomous, teleoperable mobile robots capable of performing specific reconnaissance behaviors that are currently useful to military warfighters in the near term. These behaviors may include, for example, Route Reconnaissance (in which the mobile robot is provided with one or more waypoints to be traversed while recording a map of the mobile robot's environment as it travels the path defined by the waypoints); Perimeter Reconnaissance (in which the mobile robot circumnavigates a building or other structure by identifying and following the walls of the building, while recording or transmitting a map of the terrain along the perimeter of the building); and Street Reconnaissance (in which the mobile robot travels a particular distance down a road or street, and returns to its initial location, while recording a map of the road).
• As such, instead of necessarily using excessive computational or sensory resources to solve the general problems of complex machine vision and navigation in robots, an alternative approach (as discussed herein) instead considers a narrow, but deep, focus on useful urban reconnaissance tasks.
• In view of the above, a mobile robot is discussed herein which is capable of autonomously performing navigational tasks and/or other functions. The mobile robot may perform, for example, perimeter-tracking and/or street traversal reconnaissance in autonomous or semi-autonomous operation, inter alia.
• Such an autonomous mobile robot may use range-finding or proximity sensors, optical sensors or machine vision techniques to identify linear features in the environment, such as roads, walls, parked automobiles, or buildings, for example.
• In addition, in accordance with one non-limiting example, Scaled Vector Field Histogram (SVFH) techniques may be used to detect obstacles positioned along the mobile robot's path. SVFH may also be used to provide evasive maneuvering for steering around such obstacles and reaching a clear heading, inter alia.
• In accordance with another aspect, a mobile robot may be positioned along a street or roadway, determine the general direction of the street using range-finding sensors and Hough transform techniques, and then navigate along the street out to a particular distance from its starting location, for example. The mobile robot may then turn back and return to its starting location, and may transmit map or other reconnaissance data gathered during its traversal of the street, while traversing the street.
• Thus, in accordance with various aspects, by tailoring the defined task of the mobile robot to useful, yet readily definable goals, the mobile robot can be constructed without necessarily including costly levels of sensor sensitivity and/or computational, memory or bandwidth capacity, for example. Rather, the mobile robot may be equipped with a one or more range-finding sensors such as LIDAR, radar, or sonar ranging sensors, and preferably also includes a stereo vision system, global positioning satellite receiver, inertial navigation system, odometer, inter alia.
• The obstacle avoidance system enables the mobile robot to avoid collisions with a wide range of obstacles in both outdoor and indoor environments. This system may combine range data from a 360-degree planar LIDAR system with three-dimensional (3D) depth map data provided by a stereo vision system, the combination of which is then analyzed using a Scaled Vector Field Histogram algorithm, for example. Also, a linear feature identification process may be carried out using a Hough transform algorithm applied to the range data. Data from the range sensors, obstacle avoidance, and the Hough transform may be transmitted to a a teleoperation console that presents the information to the operator graphically and in real-time, for example.
BRIEF DESCRIPTION OF THE DRAWINGS
• FIG. 1 is a profile view of a mobile robot in accordance with the present discussion; and
• FIG. 2 is an oblique overhead view of the mobile robot;
• FIG. 3 is a diagram illustrating an example of a computer-based teleoperation console and a removable media device being transferred from a mobile robot to the teleoperation console;
• FIG. 4 is a direct view of an alternative teleoperation console;
• FIG. 5 is a direct view of another alternative teleoperation console;
• FIGS. 6A,6B and6C illustrate a mobile robot surmounting a curb;
• FIG. 7 is a diagram illustrating an example hardware organization of the mobile robot;
• FIG. 8 is a plan view of a mobile robot performing an autonomous route-following operation;
• FIG. 9 is a composite plan view of a structure and a mobile robot at several locations around the structure, illustrating the performance of an autonomous perimeter-following operation;
• FIG. 10 is a plan view of a network of streets and three waypoints located at various intersections, illustrating the path traveled by a mobile robot when performing a waypoint-following operation;
• FIG. 11 is a software organization diagram illustrating an example software organization for the mobile robot;
• FIG. 12 is a software organization diagram illustrating information flow between a Hough transform software component and three reconnaissance operation software modules;
• FIG. 13 is a plan view of a semi-urban building and the environment surrounding the building;
• FIG. 14 is a example terrain map generated by a mobile robot that performed a follow-perimeter operation around the building ofFIG. 13;
• FIG. 15 is a plan view of a mobile robot navigating a corridor within a building, in which a door is closed on the right side wall;
• FIG. 16 is an example plan-view occupancy grid map generated by the mobile robot navigating the corridor as shown inFIG. 15, illustrating SVFH clear-path bins extending toward the obstacle-free paths available along the length of the corridor;
• FIG. 17 is an example plan-view occupancy grid map as illustrated inFIG. 16, further showing a strongest-line identified by an SVFH routine on the mobile robot;
• FIG. 18 is a plan view map of the mobile robot and corridor ofFIG. 15, illustrating SVFH blocked-path lines extending toward the directions surrounding the mobile robot that are blocked by obstacles;
• FIG. 19 is a plan view of a mobile robot navigating the corrider shown inFIG. 15, but with the opposite heading, and in which the door in the hallway is open;
• FIG. 20 is an example plan-view map corresponding to the scene illustrated inFIG. 19, and showing the SVFH blocked-path bins used by the mobile robot to identify walls or other obstacles that should be avoided;
• FIG. 21 is a an example plan-view occupancy grid map corresponding to the scene illustrated inFIG. 19, and showing the SVFH clear-path bins used by the mobile robot to determine a clear trajectory;
• FIG. 22 is an example plan-view occupancy grid map as illustrated inFIG. 21, further showing a strongest-line identified by an SVFH routine on the mobile robot;
• FIG. 23 is a plan view of a street, illustrating a mobile robot performing a follow-street operation in the presence of cars and trucks parked along the sides of the street;
• FIG. 24 is an occupancy grid map corresponding to the scene shown inFIG. 23, illustrating the SVFH clear-path bins identified by the mobile robot navigating the street;
• FIG. 25 is an example plan-view occupancy grid map as illustrated inFIG. 24, further showing a right and a left strongest-line identified by an SVFH routine on the mobile robot when performing the follow-street operation;
• FIG. 26 is a software organization diagram illustrating an alternative software organization for the mobile robot, which includes an arbiter routine and a waypoint navigation routine;
• FIG. 27 is an example occupancy grid map illustrating a large number of strong-line candidates identified by a Hough transform process executed on a mobile robot navigating a corridor;
• FIG. 28 is a an occupancy grid map corresponding toFIG. 27, further illustrating one strongest-line selected by the Hough transform process of the mobile robot;
• FIG. 29 is an example screen shot of a computer-based teleoperation console presenting a graphical user interface to the operator and displaying a video feed, LIDAR map, robot sensor monitor, and command line interface;
• FIG. 30 is a 3D depth map provided by a 3D stereo vision system corresponding to the view shown in the upper left corner ofFIG. 29;
• FIG. 31 is another example screen shot from a computer-based teleoperation console presenting a graphical user interface to the operator and displaying a video feed, LIDAR map, occupancy grid map, robot sensor readout, and command line interface;
• FIG. 32 is an example plan-view of a map generated by a mobile robot performing a follow-perimeter operation around the building shown inFIG. 13 using only uncorrected odometric positional information;
• FIG. 33 is another example occupancy grid map corresponding to the building shown inFIG. 13, generated by a mobile robot integrating positional information from both odometry and also a global positioning satellite system;
• FIG. 34 is a flow chart illustrating a method for controlling a robot to perform autonomous waypoint navigation;
• FIG. 35 is a flow chart illustrating a method for controlling a robot to perform an autonomous follow-perimeter operation;
• FIG. 36 is a flow chart illustrating a method for controlling a mobile robot to autonomously perform LIDAR mapping of the mobile robot's surroundings as the mobile robot traverses terrain;
• FIG. 37 is a flow chart illustrating a method for controlling a mobile robot to autonomously perform a follow-street operation;
• FIG. 38 is a flow chart illustrating a distance-based “leash” method for controlling a mobile robot to terminate an autonomous navigation operation when the distance of the mobile robot from its position at the beginning of the operation reaches a threshold;
• FIG. 39 is a flow chart illustrating a time-based “leash” method for controlling a mobile robot to terminate an autonomous navigation operation when a threshold period of time has elapsed;
• FIG. 40 is a flow chart illustrating a follow-perimeter method that employs accumulator bins to smoothly follow the perimeter of a structure, even when sharp corners may be present;
• FIG. 41 is a diagram illustrating parts of a team robot control system;
• FIG. 42 is a block diagram of a team robot control system communication scheme;
• FIG. 43 is a block diagram of conceptual building blocks of the team robot control system; and
• FIG. 44 is an example screen shot of an OCU showing path trails for two team robots in relation to a map;
• FIGS. 45A,45B and45C are example screen shots of an OCU showing a progression of improving localization/position estimates for a mobile robot as sensor input improves; and
• FIGS. 46 and 47 are example screen shots of an OCU in a team robot control system.
DETAILED DESCRIPTION
• FIGS. 1 and 2 illustrate amobile robot10 that can autonomously navigate urban environments, while generating a map of the area around the path traveled by themobile robot10, in accordance with a first embodiment. Themobile robot10 includes a range-findingsystem121 that can scan the environment surrounding the mobile robot and detect reflective surfaces of objects such as walls, automobiles, trees, furniture or the like. This range-finding data is used to generate and maintain an occupancy grid map of the mobile robot's environment along the mobile robot's path of travel, which can be conveyed to the operator of themobile robot10. Furthermore, by identifying patterns in the range-finding data corresponding to linear arrangements or features, structures such as streets, fences, or rectangular buildings may be used as navigational landmarks for orienting the mobile robot's direction of travel.
• Robot Hardware
• Themobile robot10 includes acontrol unit140 having an onboard processor for executing control software, processing sensor input and commands from an operator, and controlling the components and subsystems of themobile robot10. In accordance with one embodiment, the control software includes sets of computer software corresponding to various robot behaviors and routines, and also include operation routines that orchestrate high-level missions or operations in response to commands received from the operator. For example, the control software may include routines for a follow-perimeter operation, a follow-street operation, and a follow-waypoints operation that can be selected by the operator on a teleoperation console, as well as various concurrent behaviors or routines such as an obstacle avoidance behavior or a stasis detection behavior that function automatically during operation of themobile robot10.
• Thechassis101 of themobile robot10 may include a payload bay into which theprocessor140 or other components, such as thestereo vision system125 orrange finder121, may be detachably installed.
• FIG. 7 illustrates a computer hardware organization for themobile robot10, in which theprocessor140 exchanges data with various peripheral devices and arbitrates communication between them. In this example, the teleoperation transceiver permits theprocessor140 to receive commands from theteleoperation console21 or transmit map data, inter alia, to theteleoperation console21 during a reconnaissance operation. A peripheral interface may include a USB port into which a USB memory stick may be placed, and onto which themobile robot10 can record a map for later manual retrieval by the operator.
• Reconnaissance Operation “Leash” Constraints
• In order to ensure that themobile robot10 does not stray too far from its intended reconnaissance target or continue indefinitely, a “leash” constraint may be established for the robot operations, which ensures that the missions performed by the mobile robot have clearly defined limits or parameters. For example, a distance leash of 200 meters may be specified by the operator prior to initiating a follow-street operation, such that the mobile robot will halt its outgoing autonomous navigation down a street once it has traveled 200 meters away from its starting position, and return to the starting point (see, for example, the distance leash method illustrated inFIG. 38). Alternatively, the leash may be a time leash, in which the robot proceeds with an operation for the duration of a period of time specified by the operator, and then halts and returns once the period of time has elapsed (see, for example, the time leash method illustrated inFIG. 39).
• By defining mission parameters using a leash, the operator of the mobile robot is assured that themobile robot10 will return to its starting position regardless of whether other “end conditions” trigger the conclusion of an operation. Accordingly, themobile robot10 is of greater usefulness because of this operation-constraining capability.
• Alternatively, in accordance with other embodiments, the operator may elect not to specify a leash, or even to override an inherent leash included in the control software.
• When operating autonomously, themobile robot10 performs a mapping behavior that generates and updates the occupancy grid map. Once generated, the occupancy grid map can be transmitted to ateleoperation console21 by any appropriate mode of communication, such as Wireless Ethernet, or via a tether connection such as a USB cable between themobile robot10 and theteleoperation console21.
• Missions/Operations
• In order to perform useful reconnaissance, in accordance with one embodiment, themobile robot10 is physically light enough to be carried and transported by hand to a starting location. The operator can quickly initiate one of the various autonomous missions or operations that themobile robot10 can perform, such as follow-street, by pushing a button or switch on thechassis101 of themobile robot10 or by issuing a command from theteleoperation console21. Alternatively, for example, the operator can manually steer themobile robot10 to a starting position using theteleoperation console21 to remotely control themobile robot10, and then initiate the autonomous operation by entering a command on the teleoperation console21 (seeFIG. 29, which illustrates a command to start a follow-waypoints operation being entered into acommand line interface2522 of the teleoperation console21).
• Follow-Street
• FIG. 8 illustrates amobile robot10 performing a follow-street operation on a roadway having parkedcars961 on the side of the road. Themobile robot10 in this example uses LIDAR to detect the parkedcars961 as well as both the right and left edges of the roadway, and selects a linear path corresponding to the direction of the road by performing Hough transform analysis of the LIDAR range-finding data. The follow-street behavior attempts to find a respective Hough line on both the right and left sides of the street, and selects a heading for themobile robot10 corresponding to an average of the detected right and left Hough lines.
• FIG. 37 illustrates a method for performing the follow-street routine, in which aninitial step3701 performs a Hough transform on the data provided from therange finder121 to identify two Hough strongest lines—one on the right side and one on the left side of themobile robot10. Atstep3702 the routine determines whether a Hough strongest line has been identified on the right side; if so, the routine proceeds to step3702, while if not, the routine proceeds to step3707. Bothsteps3703 and3707 then make a similar determination whether a Hough strongest line has been identified on the left side.
• Further regarding the follow-street operation illustrated inFIG. 37, if the results ofsteps3702 and3707 are both “no”—that is, no Hough strongest line was identified for either the left or the right side—then the routine loops back to theinitial step3701 without altering the heading or velocity of the mobile robot10 (and resultingly, themobile robot10 will continue forward at whatever velocity and heading it had before that iteration of the follow-street routine). On the other hand, if a Hough strongest line was identified for only the left side, then step3708 sets the target heading for themobile robot10 to the strongest line on the left side; while if only the right side had a Hough strongest line,step3704 sets the target heading to the strongest line on the right side. If respective strongest lines were identified for both the right and left sides, then step3705 calculates an average vector based on the right and left Hough strongest lines and establishes this average as the target heading for themobile robot10. Then, after any ofsteps3704,3708 or3706, the routine steers themobile robot10 toward alignment with the target heading atstep3706 before looping back to to step3701.
• The follow-street operation typically is performed in combination with a distance leash routine that specifies the threshold distance to which the robot should continue down the street. As illustrated inFIG. 38, the distance leash routine may execute on theprocessor140 concurrently with the follow-street operation. At aninitial step3801, the distance leash routine checks whether the leash routine is active (by, for example, checking a boolean flag in the memory store of the processor140) and, if not, enters a waiting state atstep3807. If the distance leash routine is active, however, the routine proceeds to check whether the start position has been recorded atstep3802—if not, then this is the first iteration of the distance leash routine, and the routine records the mobile robot's current position as the starting position atstep3806 before returning tostep3801.
• The distance leash routine then reckons the current position of themobile robot10 atstep3803 based on localization or positional data supplied from the GPS, INS and/or odometric systems, calculates the mobile robot's distance from the starting position atstep3804, and determines whether themobile robot10 has traveled beyond the threshold distance atstep3805. If the mobile robot's distance from the starting position is less than the distance threshold, the routine returns to step3801. On the other hand, when themobile robot10 reaches the threshold distance away from the starting position, the distance leash routine terminates the follow-street operation and returns themobile robot10 back to the starting position, by automatically activating a follow-waypoints operation using only the coordinates of the starting position as the sole waypoint atstep3808.
• Similarly, a time leash may be used that operates similarly to the distance leash, but which tracks the time elapsed since a starting time of the operation, instead of distance traversed by themobile robot10. As illustrated inFIG. 39, each ofsteps3901 through3908 may be substantially similar to the corresponding steps exemplified in the distance leash method ofFIG. 38, but in which timestamps are substituted for positional coordinates.
• The follow-street behavior uses the Hough transform to detect linear features in the 360-degree planar LIDAR range data, in accordance with one alternative embodiment. Each line hypothesis with at least a minimum number of points is then classified based on whether it is on the right or left side of themobile robot10, in which
• $\mathrm{side}\left(L\right)=\{\begin{array}{cc}\mathrm{left}& \mathrm{if}{\theta }_{\mathrm{left}\min }<{\theta }_L<{\theta }_{\mathrm{left}\max }\\ \mathrm{right}& \mathrm{if}{\theta }_{\mathrm{right}\min }<{\theta }_L<{\theta }_{\mathrm{right}\max }\\ \mathrm{none}& \mathrm{otherwise},\end{array}$
• where L is a line, side(L) is the side of line L, θ_Lis the orientation of line L, θ_{left min}and θ_{left max}bracket the region of interest on the left side and θ_{right min}and θ_{right max}do the same for the right side. Currently θ_{left min}=0, θ_{left max}=θ_{right min}=π, and θ_{right max}=2π, so all lines except those orthogonal to the robot's current heading are classified as being on the left or right.
• The line headings are used to update separate accumulator arrays for the left and right sides of the robot. As before, these accumulator arrays are used to filter out transient lines generated by the Hough transform and produce a more stable desired heading.
• The value of the accumulator bins at time t is given by:
• $a_{\mathrm{left},i,t}=\left(1-\lambda \right)a_{\mathrm{left},i,t-1}+\lambda \sum _{\forall j:i\beta <\theta \left(H_j\right)<\left(i+1\right)\beta ,\mathrm{side}\left(H_j\right)=\mathrm{left}}v\left(H_j\right)$$a_{\mathrm{right},i,t}=\left(1-\lambda \right)a_{\mathrm{right},i,t-1}+\lambda \sum _{\forall j:i\beta <\theta \left(H_j\right)<\left(i+1\right)\beta ,\mathrm{side}\left(H_j\right)=\mathrm{right}}v\left(H_j\right),$
• where a_left,i,t-1is the left accumulator bin value at the previous timestep, a_right,i,t-1is the right accumulator bin value at the previous timestep, λ is the decay rate (between 0 and 1), H is the set of lines detected by the Hough transform, H_jis the jth line from this set, ν(Hj) is the number of points voting for this line, θ(Hj) is the orientation of the line, and β is the bin size. Note that, as before, all of these orientations are in world coordinates, not robot-relative coordinates (although any suitable alternative coordinate system may be utilized, such as compartmentalized regions, polar coordinates, or any other such coordinate scheme, for example).
• The selected heading corresponding to the maximum bin in each accumulator is given by:
• 
  θ_left=(i+0.5)β:∀j:a_left,i,t≧a_left,j,t
• 
  θ_right=(i+0.5)β:∀j:a_right,i,t≧a_right,j,t
• The behavior then computes the average of the left and right headings as defined by:
• ${\theta }_{\mathrm{desired}}={\theta }_{\mathrm{left}}+\frac{\Delta \left({\theta }_{\mathrm{left}},{\theta }_{\mathrm{right}}\right)}{2}$$\Delta \left({\theta }_{\mathrm{left}},{\theta }_{\mathrm{right}}\right)=\{\begin{array}{cc}{\theta }_{\mathrm{right}}-{\theta }_{\mathrm{left}}& \mathrm{if}-\pi <{\theta }_{\mathrm{right}}-{\theta }_{\mathrm{left}}\le \pi \\ {\theta }_{\mathrm{left}}-{\theta }_{\mathrm{right}}& \mathrm{otherwise}.\end{array}$
• Follow-street then sends θ_desiredas the desired heading to the SVFH obstacle avoidance behavior. If follow-street is only able to detect strong lines (with at least a minimum number of points) on one side of the robot, it attempts to align itself with the strongest line. If follow-street is unable to detect any lines on either side of the robot, it sends a command to SVFH to move straight forward. In all cases, SVFH avoids obstacles and attempts to drive the robot along the closest open direction to the desired heading.
• Follow-Perimeter
• FIG. 9 illustrates amobile robot10 performing a follow-perimeter operation around arectangular building3, in which themobile robot10 is shown at three locations along the traversed path around thebuilding3. During a follow-perimeter operation, the robot uses Hough transform analysis of the LIDAR data, similar to the follow-street behavior, but also employs a heading-voting method to smoothly and correctly navigate around corners, avoiding unstable turning behavior that could otherwise occur when themobile robot10 encounters a sharp corner. Furthermore, the follow-perimeter routine may record the coordinates of its starting position when a perimeter-following operation is initiated, and determine whether themobile robot10 has completed a circuit around the navigation target by periodically comparing the current position of themobile robot10 to the starting position.
• FIG. 35 also illustrates a method for performing one aspect of a follow-perimeter routine, in which themobile robot10 approaches the Hough strongest line until it reaches a specified distance, at which point themobile robot10 proceeds in alignment with the Hough strongest line. Atstep3501, the Hough transform is performed to identify the strongest line based on the data provided by therange finder121, atstep3502. Atstep3503, it is determined whether the current distance from the Hough strongest line is at the desired distance; if not, the routine steers toward the Hough strongest line atstep3504, proceeds forward for a distance atstep3505, and then loops back tostep3501. Otherwise, atstep3506, the routine determines whether the mobile robot's current heading is aligned with the Hough strongest line. If so, the routine returns to step3501; but if not, the routine steers toward alignment with the Hough strongest line atstep3507 before looping back tostep3506.
• Also, with regard to additional aspects of the follow-perimeter operation,FIG. 40 illustrates a method in which the follow-perimeter routine reckons the mobile robot's current position atstep4001, and determines whether a starting position has already been recorded atstep4002. If not, step4007 records the coordinates of the mobile robot's current position as the starting position, and loops back tostep4001. Otherwise,step4003 checks whether the current position is within the threshold distance of the starting position; and, if so, halts the follow-perimeter operation atstep4008 because it has completed a circuit around the perimeter of the navigation target.
• Further in view ofFIG. 40, if the result ofstep4003 is “no,” the follow-perimeter routine then performs a Hough transform on the occupancy grid map atstep4004 to ascertain a strongest line to follow, and updates the accumulator array with the current Hough strongest line atstep4005. Then,step4006 outputs a target heading based on the updated accumulator array, before returning tostep4001.
• In order to ensure smooth following around the perimeter of a navigation target, the follow-perimeter behavior generates and continuously updates an accumulator array of target headings over time, wherein at each iteration of the routine, the effect of older target headings to the accumulator array decays until they no longer effect the currently generated target heading. Nonetheless, the hysteresis effect of more recent previous headings on the accumulator array dampen any sudden shifts in the current target heading relative to the immediately previous target heading, so that unstable steering caused by oscillation is avoided even when themobile robot10 encounters a sharp corner along the perimeter being traversed.
• As an example, the follow-perimeter routine may select one of 72 bins each corresponding to 5 degrees, among the full possible range of 360 degrees of orientation. The value of an accumulator bin a_iat time t is then given by:
• 
  a_i,t=(1−λ)a_i,t-1+λΣ_{∀j:iβ<θ(Hj)<(i+1)β}v(H_j),
• in which a_i,t-1is the accumulator bin value at the previous timestep, λ is the decay rate (between 0 and 1), H is the set of lines detected by the Hough transform, H_jis the jth line from this set, ν(Hj) is the number of points voting for this line, q(Hj) is the orientation of the line, and β is the bin size, and in which the coordinates are all provided as global coordinates, rather than robot-relative coordinates. By continuously updating the accumulator bins using an algorithm based on this equation, steering and perimeter tracking is improved, while steering oscillation is reduced.
• The follow-perimeter behavior may generate a desired heading based on the relative orientation and desired range to the tracked wall. For example, for left wall following,
• $\theta ={\theta }_w+k\left(r_w-r_d\right)\mathrm{if}k\left(r_w-r_d\right)<\frac{\mathrm{<figure-callout\; label="\pi "\; filenames="US20110208357A1-20110825-D00003.png,US20110208357A1-20110825-D00040.png"\; state="\{\{state\}\}">\pi </figure-callout>}}{4},\theta ={\theta }_w+\frac{\mathrm{<figure-callout\; label="\pi "\; filenames="US20110208357A1-20110825-D00003.png,US20110208357A1-20110825-D00040.png"\; state="\{\{state\}\}">\pi </figure-callout>}}{4}\mathrm{if}k\left(r_w-r_d\right)\ge \frac{\mathrm{<figure-callout\; label="\pi "\; filenames="US20110208357A1-20110825-D00003.png,US20110208357A1-20110825-D00040.png"\; state="\{\{state\}\}">\pi </figure-callout>}}{4}.$
• For right wall following,
• $\theta ={\theta }_w-k\left(r_w-r_d\right)\mathrm{if}k\left(r_w-r_d\right)<\frac{\mathrm{<figure-callout\; label="\pi "\; filenames="US20110208357A1-20110825-D00003.png,US20110208357A1-20110825-D00040.png"\; state="\{\{state\}\}">\pi </figure-callout>}}{4},\theta ={\theta }_w-\frac{\mathrm{<figure-callout\; label="\pi "\; filenames="US20110208357A1-20110825-D00003.png,US20110208357A1-20110825-D00040.png"\; state="\{\{state\}\}">\pi </figure-callout>}}{4}\mathrm{if}k\left(r_w-r_d\right)\ge \frac{\mathrm{<figure-callout\; label="\pi "\; filenames="US20110208357A1-20110825-D00003.png,US20110208357A1-20110825-D00040.png"\; state="\{\{state\}\}">\pi </figure-callout>}}{4},$
• where θ is the behavior's desired heading in radians (relative to the robot's current heading), θ_wis the orientation of the wall in radians (relative to the robot's current heading), r_wis the range to the wall in meters, r_dis the desired range to the wall in meters, and k is a constant (for example, π/8).
• This desired heading may then be passed to the SVFH obstacle avoidance behavior. The SVFH behavior then selects the obstacle-free heading that is closest to the desired heading output by follow-perimeter. This allows themobile robot10 to reactively steer around obstacles that are located next to walls, and then resume wall-following automatically when the obstacle is no longer present.
• Follow-Waypoints
• FIG. 10 illustrates a network of streets and intersections being traversed by amobile robot10 performing a follow-waypoints operation. Threewaypoints1,2 and3 are located at three different street intersections, and the path the mobile robot will traverse to reach each waypoint in succession is shown by the double dotted arrow.
• Further,FIG. 34 illustrates a method for performing a follow-waypoints navigation routine, in which one or more waypoints defining an itinerary are supplied to the routine. For example, data representing the coordinates of each waypoint along the itinerary may be stored sequentially in an array or other suitable data structure in the memory store of theprocessor140, such as a queue or linked list, which the follow-waypoints routine can then access. As the waypoints are each retrieved from the memory store and processed by the follow-waypoints routine in the appropriate order, they are removed from the set of remaining waypoints one by one until the robot reaches the final waypoint in the itinerary, at which time the robot halts the follow-waypoints operation.
• When navigating toward each waypoint, themobile robot10 may identify linear features in its environment, such as streets or buildings, and follow them toward the waypoint. Further, themobile robot10 may consider a waypoint as “reached” when themobile robot10 moves within a threshold distance (for a non-limiting example, the threshold distance may be set to a radius of ten meters around the precise coordinates of the waypoint, or any other suitable distance) of the waypoint, improving operational efficiency and minimizing the possible effects of mapping or localization errors.
• When an iteration of the routine is executed,step3401 initially checks whether any waypoints remain to be processed and if not, the routine has achieved its purpose (there being no further waypoints left to process) and the follow-waypoints operation halts atstep3408. Otherwise, the coordinates of the next remaining waypoint are retrieved, removed from the set of remaining waypoints, and used as the current target waypoint atstep3402.Step3403 determines the coordinates of the current position of themobile robot10 based on data from the localization system (such as the GPS, INS, and/or odometry systems), and step3404 correspondingly generates a target heading toward the target waypoint from the current position of themobile robot10. At step3405 a Hough transform is performed on the data from therange finder121 to identify a strongest line to be used as a path to follow toward the target waypoint atstep3406.Step3407 determines whether the distance from themobile robot10 to the target waypoint is less than the threshold distance: if so, then the current target waypoint is considered “achieved” and the routine loops back tostep3401; if not, then the routine instead loops back to step3403 to continue seeking the current target waypoint.
• Mapping
• When themobile robot10 navigates through terrain in order to perform reconnaissance, the mapping behavior may automatically run concurrently with other behaviors in order to generate and transmit a map of the traversed terrain.
• FIG. 36 illustrates a mapping method for controlling themobile robot10 to generate and transmit maps of the terrain themobile robot10 passes through based on data from therange finder121. Atstep3601, an occupancy grid map is initiliazed and established in the memory store of theprocessor140; and atstep3602, the robot's current location is ascertained. At step3603 a frame of range finder data is used to update the occupancy grid map.
• Atstep3604, the mapping routine determines whether the map should be transmitted in a broadcast manner; if so,step3605 then broadcasts the map data to theteleoperation console21 and proceeds to step3606, which determines whether communication is currently possible, by any method (for example, by a secure WiFi link or a USB cable connection), with theteleoperation console21. If so, then step3607 sends the map to the teleoperation before proceeding to step3608. Atstep3608 the routine determines whether a detachable storage medium is accessible; and, if so, the routine records the map to the storage medium atstep3609 before returning tostep3603.
• The autonomous navigation behaviors, other than the waypoints operation, do not necessarily rely on any estimate of the robot's absolute position in order to navigate through their environments. Rather, the reactive follow-perimeter behavior may operate directly off the Hough transform estimates of the position of nearby walls relative to the robot, without the use of any absolute position information, for example. However, even more accurate localization may be obtained to build accurate maps of the environment.
• For example, a hybrid compass/odometry localization technique may be used, in which the compass is used to determine the robot's orientation, and odometry is used to determine the distance translated between updates. The robot's new position may be determined using the following equations:
• 
  Δ_t=√{square root over ((x_t−x_t-1)²+(y_t−y_t-1)²)}{square root over ((x_t−x_t-1)²+(y_t−y_t-1)²)}
• 
  x′_t=Δ_tcos θ_t,
• 
  y′_t=Δ_tsin θ_t
• where (x_t, y_t) is the odometry position at time t, θ, is the compass heading at time t, Δ_tis the distance traversed between time t-1 and time t, and (x′_t, y′_t) is the hybrid compass/odometry position estimate for time t.
• Pure odometry may tend to rapidly accumulate error in the estimate of the robot's orientation, while hybrid data integrated from multiple localization systems (such as a GPS, INS, or compass-tracking system) can provide highly accurate maps because of the significant improvement in localization precision.
• As examples of the difference in accuracy between localization based only on odometry versus hybrid localization integrating data from odometric in combination with GPS, INS or other such positioning systems,FIGS. 32 and 33 are both maps generated by amobile robot10 circumnavigating the same building (specifically, thebuilding963 shown inFIG. 13).FIG. 32 was generated by a map routine using only odometric localization data, and demonstrates a significant warping of the resulting map, in which thebuilding963 appears to have a triangular outline, quite different from the generally rectangular outline shown by thebuilding963 inFIG. 13. On the other hand, FIG.33—which was generated based on both odometric GPS localization data—shows a clear rectangular image that corresponds more closely to the actual outline of thebuilding963.
• In contrast, use of a compass can enable the mobile robot to reliably determine the robot's position to within a few degrees, and the hybrid compass/odometry localization method may be able to determine the robot's position accurately to within a few meters throughout the perimeter reconnaissance.
• Teleoperation Console
• As illustrated inFIG. 3, for example, theteleoperation console21 may include a computer system having araster display261 for presenting the occupancy grid map to the operator, as well as input systems such as akeyboard251,mouse252 andjoystick253. The occupancy grid map may be transmitted wirelessly from theteleoperation antenna131 of themobile robot10 via theantenna239 of theteleoperation console21. Alternatively, themobile robot10 may store the occupancy grid map on a detachable memory storage device142 (which may be a USB memory stick, a Flash RAM or SD/MMC memory chip, or the like) that the operator can retrieve when the mobile robot completes an autonomous operation and access using theteleoperation console21 or other suitable device.
• FIGS. 4 and 5 illustrate two alternative teleoperation console implementations. InFIG. 4, a walkie talkie-style teleoperation console22 includes anumeric keypad251 for entering numeric codes corresponding to robot operation commands, and aspeaker265 that can output audible tones or messages from the mobile robot10 (which may indicate operation status such as “operation complete” or “robot is stuck,” inter alia, via distinctive tones or synthesized speech). The walkie talkie can be tuned to a radio frequency used to communicate with the mobile robot, and the codes entered by the operator onto thekeypad251 can be encoded as dual-tone multifrequency (DTMF) signals recognizable by themobile robot10 and transmitted to themobile robot10 over a radio channel (as a non-limiting example, entering a three-digit numeric sequence such as “111” or “456” into thekeypad251 may command themobile robot10 to perform an operation such as a follow-perimeter or follow-street operation that is assigned to the entered sequence, inter alia).
• FIG. 5 illustrates a remote control-type teleoperation console23, which includes X-axis and Y-axis joysticks256 for directly steering themobile robot10, as well as mission-selection buttons255 for choosing an autonomous operation to be performed by themobile robot10.
• Range-Finder
• The range-finding system may include a scanning light source (for example, an infrared laser that is continuously rotated so as to scan and detect reflective surfaces of objects positioned anywhere around the mobile robot) and corresponding detector or other LIDAR (light detection and ranging)system121, as shown inFIGS. 1 and 2.
• Data from the range-findingsystem121 typically includes patterns or clusters of dots, in which each dot indicates that an object was detected by the range-finder at the location corresponding to the dot (see, for example,FIG. 14, which illustrates an occupancy grid map generated by a mobile robot circumnavigating thebuilding963 shown inFIG. 13). In order to identify linear features, and to select an appropriate heading to follow among various linear features that may be identified, themobile robot10 analyzes the range-finding data using a Hough transform. As illustrated inFIG. 27, the Hough transform may identify several line candidates. Using the Hough technique, one strongest-line can be selected from among multiple line candidates, as illustrated by the Houghstrongest line1606 inFIG. 28.
• Depending on the mode or operation selected, themobile robot10 may then steer so as to proceed in alignment with the Houghstrongest line1606 from the occupancy grid map.
• Obstacle Avoidance and SVFH
• Various obstacles may be encountered lying along the path of themobile robot10 as it operates autonomously. Therefore, themobile robot10 includes an obstacle detection and avoidance behavior for identifying and evading obstacles. In accordance with one embodiment, themobile robot10 includes a 3Dstereo vision system125 that employs binocular cameras and machine vision methods for generating a depth map (seeFIG. 30) that is used together with the range-finding data from the range-finder121 to detect obstacles. Themobile robot10 may employ a scaled vector field histogram technique to integrate the obstacle data from thestereo vision system125 and range-finder121, to identify clear headings and blocked paths, and to select an appropriate heading for steering themobile robot10 around obstacles and toward the desired navigation target.
• In accordance with one embodiment, the target heading generated by the navigation behaviors (e.g., follow-street or follow-perimeter) is first passed to the SVFH obstacle avoidance behavior, which may modify the target heading in response to an obstacle detected along the target heading.
• Automatic Flipper Deployment
• In accordance with another embodiment, the mobile robot includes a pair oftreaded flippers115 positioned adjacent themain treads110 of the mobile robot's drive system, to assist in surmounting low-lying obstacles. The current supplied to the drive motors that propel thetreads110 are monitored by an ammeter, which reports the drive motor current to the processor140 (seeFIG. 7). When themobile robot10 abuts an obstacle such as a sidewalk curb that prevents forward movement (as illustrated inFIG. 6A), the drive motor current will rise as the drive motors attempt to drive thetreads110 while themobile robot10 cannot proceed because of the obstacle. The treads may then be automatically deployed by a stasis detection behavior running concurrently on theprocessor140, which monitors the drive motor current and drives the flipper motor to engage theflippers115 when the drive current exceeds a particular threshold (seeFIG. 6B). Once theflippers115 have lifted thetreads110 above the curb, themobile robot10 can then proceed over the obstacle (as illustrated inFIG. 6C).
• Hardware/Software Organization
• FIG. 11 shows data flow among system components segregated into functional groups. At the top ofFIG. 11, the sensors of themobile robot10, such as the stereo vision, LIDAR, GPS or INS systems supply information to behaviors and routines executing on theprocessor140. The drive motor current sensor, which may include an ammeter on thechassis101, supplies appropriate information to the stasis detector so that the stasis detector routine can deploy the flippers automatically when the drive motor current indicates collision with an obstacle.
• Further,FIG. 12 illustrates a detail of information flow between a Hough transform routine and the three mission routines (the follow-street, follow-perimeter and follow-waypoints routines). In addition the Hough transform information, the teleoperation transceiver may provide instructions to the mission routines, such as a distance leash to be used during a follow-street mission or a command to abort a mission, for example.
• FIG. 26 illustrates an alternative software organization for a mobile robot that navigates autonomously using waypoint navigation. The mobile robot control software includes a localization subsystem for reckoning the mobile robot's position in world coordinates based on positional data received from a GPS receiver, an odometry tracker, and an inertial navigation system. The localization information includes the robot's coordinates in the global X, Y, and Z planes, and may also specify the robot's attitude (θ) pitch and roll orientations relative to the ground plane.
• This localization information is used made available to a waypoint navigator routine, an obstacle map routine, and a stasis detector routine. Each of these routines outputs to an arbiter routine, which processes the incoming data and outputs velocity and turn rate commands to the drive system control routine for causing the drive system to appropriate steer and propel the robot toward a navigation goal.
• As discussed, the follow-waypoints routine uses the localization information to select a target trajectory for themobile robot10—for example, by comparing the robot's global coordinates to the coordinates of the next waypoint in the mission information provided to the robot prior to undertaking the waypoint navigation, and calculating the angle between the mobile robot's current heading and the next waypoint. In accordance with the software organization illustrated inFIG. 26, the target trajectory is then passed on to the arbiter routine, which converts the target trajectory information into velocity and turn rate commands suitable for the drive system control routine to control the drive system to move the mobile robot across the terrain toward the appropriate heading and with the appropriate speed.
• The mapping routine receives input from therange finder121 and thestereo vision system125, and constructs a grid occupancy map based on this input. The occupancy grid map is supplied to the arbiter or SVFH obstacle detection routine, where it is used in combination with the target trajectory to adjust the actual steering and/or velocity commands issued to the drive system control routine.
• As a result, when detectable obstacles are encountered along the path between the mobile robot and its navigation target, the arbiter or SVFH obstacle detection routine can deduce their presence and location from the occupancy grid map and alter the steering or velocity of the mobile robot so as to swerve around the obstacles. Further, the follow-waypoints routine need not receive the occupancy grid map nor take it into account, because the arbiter automatically processes the occupancy grid map and evades such obstacles when encountered, and resumes steering toward the target trajectory when no obstacles are imminent.
• In accordance with one embodiment, the stasis detection behavior routine also receives the localization information regarding the mobile robot's global coordinates and can determine whether the robot is not proceeding appropriately. For example, the stasis detector may periodically compare the mobile robot's coordinates to a previous set of coordinates from a previous time and, if the two sets of coordinates are not sufficiently distant, the routine may then supply appropriate notice to the arbiter and appropriate stasis-escape or cul-de-sac avoidance actions may then be performed.
• Hough Transform
• Hough transform techniques may be employed to detect walls and road orientations for various navigation behaviors. The Hough transform is a computer vision technique that works by transforming image point coordinates into votes in the parameter space of possible lines. Each point corresponds to a vote for all of the lines that pass through that point. By finding the strongest points in the parameter space, the Hough Transform can determine the parameterized equations for the strongest lines in the image. This library of Hough transform software routines may be integrated with the local obstacle map constructed from the laser and stereo vision range data.
• The Hough transform is able to reliably find linear features in the range image and determine their location and orientation. Using the Hough transform in both outdoor and indoor environments, a mobile robot employing the Hough transform may reliably detect exterior building walls, interior hallway walls, street curbs, and rows of parked cars, for example.
• In accordance with at least one non-limiting example, the Hough transform processes range data from the LIDAR and calculates the strongest line orientations and offsets relative to the robot's current position. This system is highly accurate and reliable in determining the location and orientation of walls indoors and shows promising levels of accuracy and reliability outdoors.
• For example,FIG. 27 shows multiple Hough line candidates for themobile robot10 when positioned in a hallway (in which theoutline1610 of themobile robot10 is positioned between the line clusters). As illustrated inFIG. 28, despite the presence of obstacles and an open doorway along the right wall, the Hough transform is able to detect the correct orientation of the right wall by selecting thestrongest line1606.
• If, for example, the robot may become tilted so that it was not parallel to the ground, the laser plane would intersect the ground. In some cases, this may generate a “false positive” (spurious) potential line that could confuse the perimeter following behavior. To deal with this problem, a range filter may be deployed, which uses the sensor data from the mobile robot's pan/tilt sensor to project the laser points into 3D (see the organization diagram ofFIG. 26, for example, illustrating the flow of pan/tilt data). Then, the points in 3D that are located below the robot (relative to the gravity vector) are removed from the laser scan before this scan is passed to the Hough transform. When the robot is tilted, the laser plane will intersect the ground at some point below the robot (assuming the robot is not directly adjacent to the ground), and so these points will have a negative z-coordinate value relative to the robot. In simple urban terrain, the mobile robot can just ignore these points. In more complex terrain, the mobile robot may be instructed to explicitly avoid these points.
• This filter can work effectively to allow the robot to ignore spurious range readings that hit the ground because of the tilt of the robot, for example. This may enable the robot to successfully follow building walls without being distracted by the spurious ground hits. In addition, when the robot traverses over curbs, this can prevent the obstacle avoidance behavior from erroneously perceiving the ground as an obstacle and undesirably turning to avoid it, for example.
• SVFH Obstacle Detection and Avoidance
• To enable the mobile robot to avoid obstacles in cluttered environments, a range analysis technique known as Scaled Vector Field Histogram (SVFH) may be used. In the standard VFH technique, an occupancy grid is created in which each “square” or “cell” of the grid is filled with a probability that an obstacle exists at that point, and a polar histogram of the obstacle locations is created, relative to the robot's current location. Individual occupancy cells are mapped to a corresponding wedge or “sector” of space in the polar histogram. Each sector corresponds to a histogram bin, and the value for each bin is equally to the sum of all the occupancy grid cell values within the sector.
• The polar histogram bin values mapped to their bearings relative to the robot's heading. A bin value threshold is used to determine whether the bearing corresponding to a specific bin is open or blocked. If the bin value is under this threshold, the corresponding direction is considered clear. If the bin value meets or exceeds this threshold, the corresponding direction is considered blocked. Once the VFH has determined which headings are open and which are blocked, the robot then picks the heading closest to its desired heading toward its target/waypoint and moves in that direction.
• A bin value threshold is used to determine whether the bearing corresponding to a specific bin is open or blocked. If the bin value is under this threshold, the corresponding direction is considered clear. If the bin value meets or exceeds this threshold, the corresponding direction is considered blocked.
• FIGS. 15 through 18 illustrate amobile robot10 navigating through a hallway, facing the direction indicated by the thick arrow inFIG. 15.Walls401 and aclosed door402 on either side of the robot are detected by therange finder121 and show up as a distinctive pattern of SVFH blocked-path bins1641 extending toward the blocked directions (seeFIG. 18), while the SVFH clear-path bins1640 extend toward the obstacle-free path directions (seeFIG. 16).
• Once the VFH has determined which headings are open and which are blocked (see, for example,FIG. 18, in which the fan of bins extending out from the right and sides of themobile robot outline1610 indicate that these directions are not open, because of the presence of walls in those directions), themobile robot10 then picks the heading closest to its desired heading toward its target/waypoint and moves in that direction.
• SVFH extends the VFH algorithm such that the occupancy values are spread across neighboring bins. That is, because an obstacle that may be easily avoided at long range may require more drastic avoidance maneuvers at short range, the bin values of the SVFH technique are updated to reflect this increased importance.
• The extent of the spread is given by:
• 
  θ=k/r,
• where k is the spread factor (0.4 in the current SVFH), r is the range reading, and θ is the spread angle in radians. For example: if k=0.4 and r=1 meter, then the spread angle is 0.4 radians (23 degrees). So a range reading at 1 meter for a bearing of 45 degrees will increment the bins from 45−23=22 degrees to 45+23=68 degrees. For a range reading of 0.5 degrees, the spread angle would be 0.8 radians (46 degrees), so a range reading at 0.5 meters will increment the bins from 45−46=−1 degrees to 45+46=91 degrees. In this way, the SVFH causes the robot to turn more sharply to avoid nearby obstacles than to avoid more distant obstacles.
• As a non-limiting example, the SVFH algorithm may be implemented on a mobile robot using 360-degree range data from the infraredlaser range finder121. The range finder preferably provides a 360-degree range scan with 2 degree resolution at 5 Hz, for example. The range values from each scan are used to compute a new SVFH. The range finder may provide range data out to, for example, 12 meters, but truncated range values (for example, out to 2 meters instead of the available 12 meters) may be used to compute the SVFH, in order to reduce complexity and computational requirements.
• FIG. 19 shows amobile robot10 positioned at an intersection of two hallways, andFIG. 20 shows example SVFH blocked-path bins when the robot arrives at the hallway intersection, as shown on a real-time teleoperation console21.Vectors1641 representing SVFH blocked-path bin values have lengths that are proportional to the value of the bin associated with the corresponding direction. Long vectors correspond to a large number of nearby range readings within the bin sector. Short vectors correspond to a small number of range readings near the limit of the range window (2 meters). If no vector is present in a given direction, this means that no obstacles are within the range window in that direction.
• FIG. 21 shows the corresponding clear directions at the intersection. SVFH clear-path bin vectors1640 point toward clear directions. If no long vector is present in a given direction, this means that this direction is blocked. The SVFH detects all four of the open passages meeting at the intersection. Wider passages allow a wider range of orientations for the robot, while narrower passages allow a more limited range of orientations.
• Outdoor Obstacle Avoidance
• In accordance with one non-limiting example, planar range data from aLIDAR range finder121 is combined with 3D range data from the stereo vision system125 (or other suitable range-detecting system, such as an optical, sonar, electrostatic or magnetic sensor, inter alia, as non-limiting examples) using the SVFH technique, implemented as a control software routine. Such a system can provide robust navigational control for a mobile robot both indoors and outdoors in a wide range of urban and natural settings.
• The avoidance system detects a wide variety of potential obstacles, such as walls (indoor and outdoor), doors, furniture, cars, trucks, trees, bushes, rocks, stairs, metal railings, and chain-link fences. Both the LIDAR and the stereo vision system are positioned so they can detect obstacles that themobile robot10 is not able of climbing. Lower obstacles such as curbs, which the mobile robot can climb, are preferably excluded from the range of these obstacle sensors and are not included in the obstacle avoidance map. This allows the obstacle avoidance system to permit themobile robot10 to simply proceed over climbable obstacles, while avoiding unsurmountable obstacles at the same time. Moreover, computational and/or memory resources are also conserved because of the reduced amount of sensor information required to be processed, for example.
• A sonar sensor may also be employed for detecting obstacles such as glass and/or narrow metal wires, for example, where such obstacles are not readily detected by other sensory devices. The combination of LIDAR, stereo vision, and sonar, for example, may provide the capability to detect virtually all of the obstacles amobile robot10 might encounter in an urban environment.
• Automatic Flipper Deployment
• In some cases, the mobile robot may encounter obstacles that are below the plane of therange finder121, for example, but are difficult to detect using the vision system125 (black asphalt curbs, for example). To assist the robot in climbing over such obstacles (such as thecurb966 shown inFIGS. 6A,6B and6C), an automatic flipper deployment behavior may be utilized. When themobile robot10 attempts to climb an obstacle, but itsmain treads110 are unable to lift themobile robot10 over the obstacle, the motor currents will rise. The flipper deployment behavior monitors these currents, and when either the left or right drive motor current exceeds a threshold (for example, 15 amps), this behavior extends theflippers115 forward to assist in surmounting the obstacle (see, for example,FIGS. 6A,6B and6C). The flippers remain deployed for a minimum period of time (as a non-limiting example, 10 to 15 seconds). When both drive motor currents drop below a second, lower threshold (2 amps, for example), the flippers are retracted back to their home position.
• Cul-de-Sac/Stasis Detection and Avoidance
• In some environments, themobile robot10 could occasionally become trapped in cul-de-sacs or other dead-end paths. The robot might, for example, follow a wall into a cul-de-sac, then turn around and start to emerge, but end up following the same wall back into the cul-de-sac again, for example.
• To prevent this, a stasis detection and avoidance behavior may be provided. This behavior remembers the recent locations of the robot and prevents the robot from getting trapped in a loop. The behavior maintains a trace of the robot's recent positions and treats each point in this trace as an obstacle, which may then be passed to the SVFH obstacle avoidance system (which then regards the traced path as any other obstacle). The robot then can steer away from its recent path and move toward unexplored space, instead.
• If the robot were navigating down a very long and narrow cul-de-sac, for example, memory of previous path points could prevent it from initially turning around. In that case, the robot can wait until the path-history memory has expired (when the path-history memory is implemented as a continuously rolling, finite-capacity memory in which previously recorded memories “fall off” the end of the memory after a period of time has passed and as newer memories are recorded, for example) and then obstacle avoidance behavior would lead it back out of the cul-de-sac. When the robot emerges, the cul-de-sac behavior would prevent it from going back into the cul-de-sac, for example.
• Furthermore, when themobile robot10 operates very close to obstacles in cluttered environments, it could get stuck on a low obstacle adjacent to the rear treads110, for example—which may occur, when themobile robot10 attempts to turn and then abuts an obstacle that was too low to be detected by the robot'sLIDAR121 andvision125 system, but too high for thetracks110 to simply slide over sideways during the robot's rotation, inter alia. This is an example of the general problem of behavioral stasis, which occurs when the robot is attempting an action, but is “stuck” and unable to move.
• To increase the general robustness and capability of the system, a general stasis-escape behavior may be utilized. This behavior detects when the robot is stuck and then attempts random (or semi-random, or pre-programmed, for example) motions until the robot becomes unstuck.
• In a non-limiting example, the stasis-escape behavior maintains a stasis level variable. This variable is increased whenever the behavior system is sending a translation or rotation command, but the robot's treads are not moving (as determined by odometry). Conversely, the stasis level is reduced whenever the robot is moving. When the stasis level exceeds a threshold, an escape action is triggered, for example. The escape action may command the robot to move at a random speed (for example, −0.25 to +0.25 m/sec) and a random rotation (for example, −1.0 to +1.0 radians/sec).
• Alternatively, for example, the robot's commanded speed (or direction of rotation of the wheels, treads, or other drive system, such as to alternative between forward and reverse) and/or steering may be caused to alternate in a rythmic pattern, such as in a manner similar to “rocking” a quagmired automobile out of a snowy parking spot or mud pit (in which the driver rythmically alternates the automobile transmission from forward to reverse and back). In such a case, the robot may take advantage of self-reinforcing oscillation such that each successive cycle “rocks” the mobile robot further and further out of its stuck position. Sensors and/or analysis routines may be employed to detect whether a certain rhythm or motion pattern is producing self-reinforcing oscillation, for example, and may change the stasis avoidance behavior to another method when no progress is detected after a particular number of attempts.
• If themobile robot10 starts moving, the stasis level begins to fall, and when it falls below threshold, the escape action is terminated, and control is returned to the robot's regular behaviors. If the robot does not start moving, then after a specified period of time (2 seconds), another random escape action is generated, and the new translation and rotation commands are sent to the robot. The stasis-escape behavior may repeat this process until the robot starts moving.
• Obstacle Classification and Visualization using 3D Stereo Vision
• In accordance with at least one non-limiting example, an obstacle classification system analyzes the set of 3D return points from the stereo vision sensor by converting the depth map information back into 3D point information. As illustrated inFIG. 30, for example, this information is then projected to a 2D grid where statistics are kept on all the points that “hit” each 2D grid cell. The algorithm keeps track of the minimum and maximum height of all points that hit a cell as well as a count of the number of points that hit a cell. This quantized processing allows for very efficient and low-latency handling of the 3D information which is suitable for OD/OA systems.
• Once the statistics grid is computed to filter out spurious readings, false short depth-readings a set of heuristic rules are used to classify grid cells as obstacles based on the properties of the robot system. These are:
• Grid-to-Grid Slope Threshold:
• • This threshold is applied to detect obstacles too steep for the robot to climb. Surfaces that appear to change at a slope >45 degrees are considered to be obstacles fro this system. This allows the strereo vision to report trees, walls and posts as obstacles, but to omit kerbs and gentle slopes from the obstacle map even if they are within range.
• Grid Minimum Height Threshold:
• • This threshold is applied to detect “overhanging” obstacles that don't touch the ground. Thus, the system won't report a high truck body as an obstacle since it can pass under the truck, but the slope threshold will report the tires as obstacles to drive between.
• The combination of these two heuristic classification approaches yield good 3D obstacle perception that is matched to the vehicle mobility characteristics. In accordance with at least one exemplary implementation, as a non-limiting example, the system including stereo processing, data transmission, obstacle classification and conversion to range data operates at 15 Hz with speed being limited by stereo processing throughput at the moment. This allows robot speeds of over1m/s with a fairly good safety margin, given the ˜5 m detection range of the stereo obstacle system.
• Conversion To Range Scan Data using 3D Stereo Vision Input
• In accordance with another non-limiting example, once the data have been classified as obstacles, these obstacles are then integrated into the SVFH map. The laser range data is used for obstacle avoidance and map building in the mobile robot's autonomous behavior modes. In addition, the laser data can provide augmented spatial awareness in tele-operated modes.
• Conversion of 3d Stereo Vision Data to Range Scan Data
• Once the data have been classified as obstacles, these obstacles are then converted into the range/bearing scan format used by the obstacle avoidance software. The mobile robot's control software can build a temporary Local Perceptual Space (LPS) map based thereon that represents the region of space near to themobile robot10. This map preferably includes a quantized point cloud that represents all recent range readings in the mobile robot's local coordinate frame.FIG. 31 shows the mobile robot operator's control unit's (OCU's) real-time graphical display of theLPS map2514.
• The pattern of points on theLPS map2514 represent recent range returns in the robot's current local coordinate frame. As the robot translates or rotates, previously detected points are transformed into the current reference frame. These points decay over time (on the order of seconds) as they are no longer detected. This means that moving obstacles can be dynamically represented in the map as their position changes in the world.
• An advantage of this map over a purely reactive approach is that recently detected obstacles can be used in path planning, even if they are no longer visible. So if the robot moves past an obstacle or around a corner, what it saw previously can be used to plan its path.
• Each robot navigation behavior (such as the follow-perimeter, follow-street or follow-waypoint behaviors, inter alia) may output a set of desired trajectories that fan out around the most desired path. For example: a behavior may want to move forward at 2 meters per second while turning right at 1 radian per second. The behavior then provides a set of adjacent trajectories at 2 m/sec and 0.6 rad/sec, 0.8 rad/sec, 1.0 rad/sec, 1.2 rad/sec, 1.4 rad/sec, et cetera.
• The obstacle detection and avoidance behavior projects these trajectory arcs through the LPS and detects potential collisions. A robot motion model including information regarding the mobile robot's physical size and shape are used to determine the template that is swept through the LPS searching for obstacle collisions.
• The trajectories are then scored based on time-to-collision (higher is better) and deviation from the optimal path (lower is better). The combined score is used to select the best turn command. In addition, the robot's speed may also be reduced to slow down for nearby obstacles.
• Group Robot Control and Integration
• In accordance with another embodiment, a team of two or moremobile robots10 may be integrally controlled so as to perform joint operations. Furthermore, a tablet computer may be used to facilitate control of the team of mobile robots, by presenting map and location information for the robots comprising the team and accepting commands through a touch-sensitive screen. As shown inFIG. 41, for example, first and secondmobile robots11,12 are controlled as a team through wireless interface with a tablet-type teleoperation console (also referred to herein as an operator control unit or OCU)24. The Tablet PC is an excellent platform as an OCU as it allows the operator to be mobile even while controlling a team of robots, and provides a simple and intuitive interface.
• The hardware and many of the software features of theteam robots11,12 may be substantially similar to any of the embodiments of themobile robot10 discussed hereinabove, for example. Also, theteam robots11,12 may include additional hardware or software features to facilitate team operation. For example, the team robot control system may include extended vector field histogram/scaled vector field histogram functionality, and may include additional behavior routines for performing “assemblage”-type group robot behaviors (for example, “formation” behaviors, a follow-the-leader behavior, a caravaning behavior, distributed landscape exploration behaviors, etc., inter alia). Using an appropriate software platform and software object model, such as the AWARE system, group sensors and behaviors may be abstracted as network services accessible to OCU client software, as a non-limiting example. The system may further include features such as visual landmark modeling or group tasking, for example.
• FIG. 46 shows an example of OCU screens for controllingteam robots11,12, in which the windows on the left and right relate torobot11 and12, respectively. Also,FIG. 47 shows a further example of OCU control screen windows for theteam robots11,12.
• As an advantage, the team control software permits eachmobile robot10 in a team a high degree of local autonomy, offloading robot-specific details such as obstacle avoidance to the onboard obstacle avoidance behavior on each robot. Accordingly, the team control system need only send high-level navigational goals to theteam robots11,12, and then eachteam robot11,12 will take care of any necessary path adjustments or obstacle avoidance issues on its own as it proceeds toward the high-level navigational goal.
• During a team navigation operation, theOCU24 may display a map of the vicinity of theteam robots11,12, and superimposed representations of the robots' respective positions in the map. For example,FIG. 44 shows a screenshot from anOCU24 controlling first andsecond team robots11,12, in which the dottedline4411 indicates the path traversed by thefirst team robot11 and thesolid line4412 shows the path traversed by thesecond team robot12. Stylizedarrows4421,4422 represent the location and orientation of theteam robots11,12 on the map. Also, by selecting one of the robots by tapping it with a stylus, and then tapping the control buttons at the bottom of the screen, the operator can command the selected robot to engage or disengage its brakes, for example.
• With these and other situational awareness representations, the operator can quickly gain an overview of the entire system with only a quick glance. If an anomaly is detected, the operator can then tunnel down into more detailed information, including full telemetry from eachteam robot11,12, to diagnose and respond to any issues.
• The system may include an interface for designating waypoints for theteam robots11,12. A complete path can be specified for each robot, by providing a number of waypoints. The interface may use a single mode of operation for simplicity, which preferably allows operation with only a stylus, for example. The interface is to click on a location the robot is either at or a waypoint the robot is already scheduled to reach and then drag to the location the robot should go to and release. Any path that was previously specified past the start waypoint/robot is erased. This interface allows for any path to be specified and modified using only the stylus. Additionally, a patrol loop can be specified by adding a path segment that ends on a previous waypoint. The robot interprets this as a designation of a loop that should be continually traversed until the robot is given other orders.
• Another interface feature may be included which permits oneteam robot11 to follow anotherteam robot12. This feature requires that bothteam robots11,12 know their current accurate position. Linking robots in this way sets up a goal for the following robot to get near the leading robot. Robots linked in this way form a leader-follower structure. Robots can be chained together in this method into a single file line, for example.
• FIG. 42 illustrates the communication organization between the constituent elements of the robot team control system, in which the operator interacts with the OCU, and the OCU communicates wirelessly with groups of team robots through a wireless LAN.
• FIG. 43 illustrates the building blocks of system organization used in the team robot control system.
• Local Perceptual Space (Local Obstacle Memory)
• A key part of the OD/OA system is the local perceptual space (LPS) which stores a representation of obstacles in the immediate vicinity of the robot. The local perceptual space is stored as an occupancy grid. The grid covers a 4 m×4 m area with 0.12 m×0.12 m cells. Each grid cell stores a simple exponentially weighted sum of evidence for/against an obstacle in that grid cell. The data in the grid cells decays exponentially (which leads to an efficient implementation) with a half life of 0.4 seconds.
• In accordance with one embodiment, the grid is centered on the robot in an efficient manner. The grid is generally oriented in the same direction which is aligned with odometric coordinates (a coordinate frame updated solely based on odometry). The robot's current position and orientation in odometric coordinates is also stored. Each grid cell covers a range of odometric coordinates. The exact coordinates covered are not fixed, however, but can change occasionally as the robot moves. The grid acts like a window into the world in the vicinity of the robot. Everything beyond the grid edges is treated as unknown. As the robot moves, the area covered by the grid also moves. The position of the robot has an associated grid cell that the robot is currently inside. This cell acts as the center of the LPS. The grid is wrapped around in both x and y directions (giving the grid a toroidal topology) to provide a space of grid cells that moves with the robot (whenever the robot crosses a cell boundary) and stays centered on the robot. Cells directly opposite from the position of the robot in this grid are ambiguous as to which direction from the robot they represent. These cells are actively cleared to erase old information and are dormant until they are no longer directly opposite from the robot. This structure provides for a fast, efficient, and constant memory space LPS.
• To use the LPS in behaviors, a virtual range scan is computed to the nearest obstacles. This virtual range scan represents what a range scanner would return based on the contents of the LPS. Converting to this form allows the same behaviors that were developed with SICK LIDAR data to also be used with data that originates from a SwissRanger which has a significantly smaller field of view.
• Heading and speed control to achieve the desired location in odometry coordinates are calculated at the same time. First, for every direction the robot could go (maximum of ˜45 degrees away from goal), the maximum safe speed in that direction is calculated. Maximum safe speed is calculated proportional to distance to closest obstacle in that direction that the robot would hit with an upper and lower bound. The heading which results in the fastest speed in the direction of the goal is chosen, i.e., the effective speed is based off of the actual speed when projected onto the goal direction.
• Localization/Mapping for Team Robot Control
• For controllingmultiple team robots11,12, it is important that the operator can communicate his intentions to the robots as easily as possible. For this communication to be effective, it is important that the operator and the robots share a common reference frame to which commands and instructions can be related. The most common reference frame is the physical world. Humans naturally build an internal map with a notion of places and the ways in which they are connected. Robots do not automatically have this ability, but they can be programmed to have this ability. The process of determining the position of a robot within a map is known as localization. When location and mapping are done at the same time, the process is known as Simultaneous Localization and Mapping (SLAM).
• By having a shared notion of a map of the geography of the environment, it makes it easy for the operator to communicate intent in a language that theteam robots11,12 can understand. For example, by having a map that is shared between operator and team robots, the operator can direct a robot to go to a particular location simply by clicking where on the map the robot should go. Having a shared map also makes it easy to express desires such as explore this area, patrol this area, follow this path, and take pictures from these locations just to name a few possible tasks.
• To enable this kind of natural communication and shared understanding, we localization capability is provided for theteam robots11,12. The algorithm used for localization may be, for example, Monte Carlo Localization. The algorithm works by maintaining a probability distribution over robot positions. At any point in time, the robot has a notion of the probability of being at a particular location and orientation. For computational efficiency reasons, the probability distribution is represented as a set of discrete guesses of possible locations that the robot might be in. These guesses are commonly called particles or samples. Each particle represents a single, exact position that the robot might be at. For example, a particle might represent the hypothesis that the robot is at exactly at (23.452, 10.024) with an angle of −45.32 degrees relative to the origin.
• As the robot moves, the particles are moved in the same fashion. So if the robot moves forward 1 meter, each particle moves forward approximately 1 meter with some error introduced to represent error in the robot's motion. As sensor readings become available, each particle is evaluated to determine how likely it is that the robot would have seen those sensor readings from the position at which the particle is located. This evaluation requires that the robot have a map of its environment. The particles are then weighted based on these likelihoods. Based on these weights, some particles are duplicated and others are removed to produce a new set of samples with uniform weights. Particles with higher weights are more likely to be duplicated and particles with lower weights are more likely to be dropped. All of these updates are done based on a probabilistic foundation which provides proof that the algorithm behaves properly under a set of assumptions, although not all of these assumptions are met in practice. In practice, the algorithm performs well in real world settings and has been extensively used and studied over the last 6 years.
• A pictorial representation of the workings of the algorithm is shown inFIGS. 45A,45B and45C. InFIG. 45A, the robot's position is initialized to an unknown position near an office. Each smallsquare particle4591 shows one possible location of the robot (the location of a particle). Thethin orientation lines4592 extending from the center of each square shows the orientation of the particle. Toward the edge of the circular region you can see theindividual particles4591. In the center figure, the robot has received more information, primarily about a wall to the robot's left. At this point, you can see that the robot has been able to eliminate many positions as possibilities, but multiple distinct possibilities remain. In the figure on the right, the robot has obtained additional information and has converged to the correct position of the robot, with theparticles4591 andorientation lines4592 all substantially in alignment.
CONCLUSION
• Although the above-noted discussion describes components and functions implemented in the embodiments with reference to particular standards and protocols, the invention is not limited to such standards and protocols. The terms “standard” or “protocol” are not limited in meaning to public or publicly promulgated concepts, and are inclusive of proprietary and private systems or rules. Standards for Internet and other packet switched network transmission, for public telephone networks, for wireless communication, for buses and interfaces, or for cellular telephone networks represent examples of the state of the art. Such standards are periodically superseded by faster, more capable, or more efficient equivalents having essentially the same functions.
• In many cases such standards coexist with similar competing and complementary standards and variations of each. Accordingly, competing and complementary standards (as well as variations of explicitly discussed standards and variations of competitor and complementary standards) having the same role in a network are considered to fall within the literal claim language, or alternatively as equivalents of the literal claim language. Superseding standards, files types, protocols, directory structures, language variations, and/or new generations of each, are also considered to fall within the literal claim language, or alternatively to be equivalents of the literal claim language.
• It should be noted that not all of the functions and features described in detail herein are necessary for a complete and functional expression of the invention. For example, in accordance with at least some example embodiments, the mobile robot can be used without a teleoperation console; likewise, the teleoperation console may be operated without a robot. Further, in various embodiments according to the present discussion, a different mobility platform may be provided for the mobile robot, or for any permutation or combination of mobile robot, teleoperation console, or other component. No one (or more) described element or feature is implicitly or inherently critical or necessary to the operation of the invention, except as explicitly described herein.

Claims (20)

1.-20. (canceled)

21. A mobile robot, comprising:

a drive system configured to propel the mobile robot across terrain;

a range sensor configured to detect a distance between the mobile robot and a physical object in an environment of the mobile robot; and

a processor communicatively connected to the range sensor and to the drive system, and configured to execute: a mapping routine configured to maintain an occupancy grid map corresponding to linear features of one or more physical objects in the environment of the mobile robot of the environment of the mobile robot, a linear feature routine configured to track votes for line hypotheses to detect one or more linear patterns in the occupancy grid map and to determine a strongest line among the one or more linear patterns, and a navigational routine configured to control the drive system to move the mobile robot in a direction aligned in parallel with the linear feature of a first physical object corresponding to the strongest line among the one or more linear patterns;

wherein the strongest line includes the line hypothesis corresponding to a linear feature of the first physical object in the environment of the mobile robot and having a greatest number of votes among the line hypotheses that have at least a threshold number of votes.

22. The mobile robot according toclaim 21, further comprising a position reckoner configured to determine a location of the mobile robot.

23. The mobile robot according toclaim 21, further comprising a teleoperation transceiver configured to receive a command from a teleoperation console and/or to transmit map data to the teleoperation console.

24. The mobile robot according toclaim 22, wherein the processor is further configured to execute a localization routine configured to update the occupancy grid map using a scaled vector field histogram based on input from the range sensor and the position reckoner.

25. The mobile robot according toclaim 21, wherein the processor is further configured to designate a location of the mobile robot when the mapping routine begins maintaining the occupancy grid map as an initial location, and

wherein the navigational routine is further configured to prevent the drive system from moving the mobile robot farther than a leash distance from the initial location.

26. The mobile robot according toclaim 23, wherein the processor is further configured to execute a perimeter-following routine configured to cause the mobile robot to circumnavigate a reconnaissance target identified using a Hough transform, to record the occupancy grid map when circumnavigating the reconnaissance target, and to transmit the recorded occupancy grid map to the teleoperation console.

27. The mobile robot according toclaim 23, wherein the processor is further configured to execute a street-following routine configured to cause the mobile robot to navigate to a first location selected by the operator, to identify a street using a scaled vector field histogram, to traverse the identified street to a specified distance from the initial location, to record the occupancy grid map when traversing the identified street, to return to the first location, and to transmit the recorded occupancy grid map to the teleoperation console.

28. The mobile robot according toclaim 25, further comprising an operator interface configured to cause the mobile robot to perform a robot mission starting from the initial location of the mobile robot when the operator interface is operated.

29. The mobile robot according toclaim 23, wherein the processor is further configured to execute a rallying routine when communication with the teleoperation console fails.

30. The mobile robot according toclaim 29, wherein the rallying routine is configured to cause the mobile robot to move toward a predetermined location until communication is established with the teleoperation console.

31. The mobile robot according toclaim 29, wherein the rallying routine is configured to cause the mobile robot to reverse its heading until communication is established with the teleoperation console.

32. A method for controlling a mobile robot, comprising:

detecting a distance between the mobile robot and a physical object in an environment of the mobile robot;

maintaining an occupancy grid map of the environment of the mobile robot,

detecting one or more linear patterns in the occupancy grid map corresponding to linear features of one or more physical objects in the environment of the mobile robot;

determining a strongest line among the one or more linear patterns; and

navigating the mobile robot in a direction aligned in parallel with the linear feature of a first physical object corresponding to the strongest line;

wherein determining the strongest line comprises executing a linear feature routine comprising tracking votes for line hypotheses to detect the one or more linear patterns in the occupancy grid map; and

wherein the strongest line includes the line hypothesis corresponding to a linear feature of the first physical object in the environment of the mobile robot and having a greatest number of votes among the line hypotheses that have at least a threshold number of votes.

33. The method according toclaim 32, further comprising:

reckoning a position of the mobile robot using a global positioning satellite receiver, an odometer, or an inertial navigation system; and

updating the occupancy grid map using a scaled vector field histogram based on the reckoned position of the mobile robot and the detected distance between the mobile robot and the object in the environment of the mobile robot.

34. The method according toclaim 33, further comprising preventing the mobile robot from navigating farther than a leash distance from an initial location.

35. The method according toclaim 32, further comprising receiving a command from a teleoperation console.

36. The method according toclaim 32, further comprising:

identifying a reconnaissance target using a Hough transform;

navigating the mobile robot to circumnavigate the reconnaissance target;

recording the occupancy grid map when circumnavigating the reconnaissance target; and

transmitting the recorded occupancy grid map to a teleoperation console.

37. The method according toclaim 32, further comprising:

navigating the mobile robot to a first location selected by an operator;

identifying a street using a scaled vector field histogram;

traversing the street to a specified distance from an initial location;

recording the occupancy grid map when traversing the street;

returning to the first location; and

transmitting the recorded occupancy grid map to a teleoperation console.

38. The method according toclaim 32, further comprising navigating the mobile robot toward a predetermined location or in a reverse heading until communication is established with a teleoperation console when communication with the teleoperation console fails.

39. The method according toclaim 32, wherein the mobile robot includes a drive system for propelling the mobile robot across terrain, a range sensor for detecting the distance between the mobile robot and the object in the environment of the mobile robot, and a processor communicatively connected to the range sensor and to the drive system for executing: a mapping routine for maintaining the occupancy grid map of the environment of the mobile robot, a linear feature routine for detecting the one or more linear patterns in the occupancy grid map and determining the strongest line among the one or more linear patterns, and a navigational routine for controlling the drive system to move the mobile robot in a direction aligned with the strongest line among the one or more linear patterns,

wherein the detecting the distance between the mobile robot and the object in the environment of the mobile robot is performed by the range sensor of the mobile robot, and

wherein the maintaining the occupancy grid map of the environment of the mobile robot, the detecting the one or more linear patterns in the occupancy grid map, the determining the strongest line among the one or more linear patterns, and the navigating the mobile robot in the direction aligned with the strongest line are performed by the processor of the mobile robot.

Images