技术领域Technical Field
本发明涉及无人艇自主避障技术领域，尤其涉及一种垃圾回收无人艇的自主避障方法、装置及相关设备。The present invention relates to the technical field of autonomous obstacle avoidance for unmanned boats, and in particular to an autonomous obstacle avoidance method, device and related equipment for an unmanned boat for garbage collection.
背景技术Background technique
水面无人艇是能在无人干预的情况下在各种复杂未知的水面环境下执行任务的新型载体，其具有体型小、智能化、自主化等优点。但目前国内的自主航行无人艇系统还不够完善，尤其是水面无人艇的避障技术上面仍未取得显著突破，因此，研究水面无人艇自主避障关键技术，对于提高水面无人艇自主智能化水平，具有重大意义。Unmanned surface boats are new carriers that can perform tasks in various complex and unknown water environments without human intervention. They have the advantages of small size, intelligence, and autonomy. However, the current domestic autonomous navigation unmanned boat system is not perfect enough, especially the obstacle avoidance technology of unmanned surface boats has not made significant breakthroughs. Therefore, research on the key technologies of autonomous obstacle avoidance of unmanned surface boats is of great significance to improving the autonomous intelligence level of unmanned surface boats.
现有的无人艇自主避障是无人艇完成各种水上任务的不可或缺的环节，也是无人艇实现智能化的关键。国内在水面无人艇避障规划技术领域的工作开展较晚，有采用人工势场法实现水面无人艇的全局避障规划，但因为容易陷入局部最优解，而有可能导致路径规划失败。还有其他船舶避碰算法，例如遗传算法、粒子群算法以及蚁群算法等，但都存在实时性差的问题。因为无法预先从样本数据中建立模型，所以在应用过程中需要重复进行路径优化搜索过程，造成大量无用的计算，且其依赖的评价函数较简单，在不同水域应用环境下的鲁棒性较差，避障效果不佳，适用范围小。The existing autonomous obstacle avoidance of unmanned boats is an indispensable part of unmanned boats to complete various water tasks, and it is also the key to the intelligentization of unmanned boats. Domestic work in the field of obstacle avoidance planning technology for surface unmanned boats started late. The artificial potential field method is used to realize the global obstacle avoidance planning of surface unmanned boats, but because it is easy to fall into the local optimal solution, it may cause path planning failure. There are other ship collision avoidance algorithms, such as genetic algorithms, particle swarm algorithms, and ant colony algorithms, but they all have the problem of poor real-time performance. Because it is impossible to build a model from sample data in advance, the path optimization search process needs to be repeated during the application process, resulting in a large number of useless calculations, and the evaluation function it relies on is relatively simple, the robustness in different water application environments is poor, the obstacle avoidance effect is not good, and the scope of application is small.
发明内容Summary of the invention
针对以上相关技术的不足，本发明提出一种定位效果好、路径规划合理、避障方便、准确性高的垃圾回收无人艇的自主避障方法、装置及相关设备。In view of the deficiencies of the above-mentioned related technologies, the present invention proposes an autonomous obstacle avoidance method, device and related equipment for a garbage recycling unmanned boat with good positioning effect, reasonable path planning, convenient obstacle avoidance and high accuracy.
为了解决上述技术问题，第一方面，本发明实施例提供了一种垃圾回收无人艇的自主避障方法，用于自主避障系统，所述自主避障系统包括导航定位模块、环境感知模块、路径规划模块和驱动模块，所述自主避障方法包括以下步骤：In order to solve the above technical problems, in the first aspect, an embodiment of the present invention provides an autonomous obstacle avoidance method for a garbage recycling unmanned boat, which is used in an autonomous obstacle avoidance system. The autonomous obstacle avoidance system includes a navigation and positioning module, an environment perception module, a path planning module and a driving module. The autonomous obstacle avoidance method includes the following steps:
通过所述导航定位模块采集卫星数据，并通过数据链将其观测值和站点坐标信息一起传送到无人艇上；The navigation and positioning module collects satellite data and transmits its observation values and site coordinate information to the unmanned boat through a data link;
通过将所述卫星数据和接收到的所述数据链进行实时载波相位差分处理，获得定位结果；Obtaining a positioning result by performing real-time carrier phase differential processing on the satellite data and the received data link;
所述环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，获得障碍物和无人艇目标点位的位置信息；The environment perception module obtains the location information of obstacles and unmanned boat target points by fusing the information measured by the millimeter wave radar and the monocular camera;
通过所述路径规划模块进行自主决策，通过所述驱动模块驱动所述无人艇达到自动规划路径和自主避障。Autonomous decision-making is performed through the path planning module, and the unmanned boat is driven by the driving module to achieve automatic path planning and autonomous obstacle avoidance.
优选的，所述通过将所述卫星数据和接收到的所述数据链进行实时载波相位差分处理，获得定位结果具体包括以下子步骤：Preferably, the step of performing real-time carrier phase differential processing on the satellite data and the received data link to obtain the positioning result specifically comprises the following sub-steps:
通过RANSAC算法在迭代过程消除数据中的相差值；Eliminate the difference values in the data in an iterative process through the RANSAC algorithm;
利用PCA法进行平面拟合；The PCA method was used for plane fitting;
获得厘米级的定位结果。Get centimeter-level positioning results.
优选的，所述RANSAC算法具体包括以下子步骤：Preferably, the RANSAC algorithm specifically includes the following sub-steps:
预设艇载传感器检测到的三维坐标数据集为Ω；The three-dimensional coordinate data set detected by the onboard sensor is preset as Ω;
从所述数据集中选取符合条件的最小样本子集，计算所述最小子集的模型参数作为初始参数；Selecting a minimum sample subset that meets the conditions from the data set, and calculating the model parameters of the minimum subset as initial parameters;
计算所述Ω与所述初始参数之间的差值；Calculating the difference between the Ω and the initial parameter;
将所述差值与预设的阈值相比较，获得对比结果，并根据所述对比结果为判断条件筛选出不符合条件的点并将其删除；Compare the difference with a preset threshold value to obtain a comparison result, and filter out points that do not meet the conditions and delete them based on the comparison result as a judgment condition;
再重复以上过程，不断迭代，最终获得一个数学模型参数。Repeat the above process and iterate continuously to finally obtain a mathematical model parameter.
优选的，所述RANSAC算法得出的结果有用的概率为p，所述p的表达式(1)如下：Preferably, the probability that the result obtained by the RANSAC algorithm is useful is p, and the expression (1) of p is as follows:
p＝1-[1-ωn]K…(1)；p＝1-[1-ωn ]K …(1);
其中，ω为数据在样本点集中的概率，n为模型拟合一次所需的坐标点的个数，K为实际的迭代次数，所述K的表达式(2)如下：Wherein, ω is the probability of the data in the sample point set, n is the number of coordinate points required for model fitting once, K is the actual number of iterations, and the expression (2) of K is as follows:
通过RANSAC算法预处理后得到点集，利用PCA法计算平面的法向量和所有采样点到目标平面的距离d，所述d的表达式(3)如下：After preprocessing with the RANSAC algorithm, the point set is obtained. The PCA method is used to calculate the plane normal vector and the distance d from all sampling points to the target plane. The expression (3) of d is as follows:
其中xi,yi,zi为采样点坐标，N为采样点个数，该xi,yi,zi分别的均值表达式(4)如下：Where xi , yi ,zi are the coordinates of the sampling points, N is the number of sampling points, and the mean expressions (4) of xi , yi ,zi are as follows:
再求出di的标准差δ，所述标准差δ的表达式(5)如下：Then find the standard deviation δ of di . The expression (5) of the standard deviation δ is as follows:
当di<2δ时，保留该点，通过不断迭代得到最佳二维拟合方程；得到无人艇周边环境信息并进行以上处理之后，使用目标检测算法yolov4对水面的垃圾进行识别，再利用KCF算法对目标垃圾进行跟踪。When di <2δ, this point is retained, and the best two-dimensional fitting equation is obtained through continuous iteration. After obtaining the surrounding environment information of the unmanned boat and performing the above processing, the target detection algorithm yolov4 is used to identify the garbage on the water surface, and then the KCF algorithm is used to track the target garbage.
优选的，所述环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，获得障碍物和无人艇目标点位的位置信息具体包括以下子步骤：Preferably, the environment perception module obtains the position information of obstacles and unmanned boat target points by fusing the information measured by the millimeter wave radar and the monocular camera, which specifically includes the following sub-steps:
获取毫米波雷达和单目摄像头所获取的环境信息，其中，所述环境信息包括所述毫米雷达的点云数据和所述单目摄像头的图像数据；Acquire environmental information acquired by the millimeter wave radar and the monocular camera, wherein the environmental information includes point cloud data of the millimeter wave radar and image data of the monocular camera;
对所述点云数据和所述图像数据进行预设时间同步处理；Performing preset time synchronization processing on the point cloud data and the image data;
对所述点云数据和所述图像数据分别进行滤波和降噪处理；Performing filtering and noise reduction processing on the point cloud data and the image data respectively;
对降噪后的所述图像数据进行目标检测；Performing target detection on the image data after noise reduction;
将所述点云数据和所述图像数据进行投影变换，再进行数据融合定位，获得所述无人艇周围三维空间下的障碍物运动信息以及所对应的坐标值。The point cloud data and the image data are projected and transformed, and then data fusion positioning is performed to obtain obstacle movement information and corresponding coordinate values in the three-dimensional space around the unmanned boat.
优选的，所述通过所述路径规划模块进行自主决策，通过所述驱动模块驱动所述无人艇达到自动规划路径和自主避障具体包括以下子步骤：Preferably, the autonomous decision-making by the path planning module and the driving module driving the unmanned boat to achieve automatic path planning and autonomous obstacle avoidance specifically include the following sub-steps:
根据所述自主决策建立预设的无人艇运行学模型；Establishing a preset unmanned boat operation model according to the autonomous decision;
根据所述无人艇运行学模型将深度学习神经网络融合进DPG策略中，获得DDPG算法；According to the unmanned boat operation model, the deep learning neural network is integrated into the DPG strategy to obtain the DDPG algorithm;
根据所述无人艇上的艇载传感器采集的艇载数据；Onboard data collected by onboard sensors on the unmanned boat;
根据所述艇载数据采用速度障碍法进行处理；Processing the boat-borne data using the speed barrier method;
根据处理获取一组样本数据；Obtain a set of sample data according to the processing;
将所述样本数据放入经验池中；Putting the sample data into an experience pool;
选择策略动作和做出评价；Select strategic actions and make evaluations;
根据所述策略动作和做出评价选择航向角度。A heading angle is selected based on the strategic actions and the evaluation made.
优选的，所述自主避障方法还包括以下子步骤：Preferably, the autonomous obstacle avoidance method further includes the following sub-steps:
所述DDPG算法做出错误选择；The DDPG algorithm makes the wrong choice;
将所述DDPG算法退回为上个状态；Return the DDPG algorithm to the previous state;
DDPG网络输出动作值并提高噪声；The DDPG network outputs action values and increases noise;
根据所述动作值控制所述无人艇执行相应动作，并得到样本数据；Controlling the unmanned boat to perform corresponding actions according to the action value, and obtaining sample data;
根据所述样本数据进行DDPG网络更新。The DDPG network is updated according to the sample data.
第二方面，本发明实施例还提供一种垃圾回收无人艇的自主避障装置，用于自主避障系统，所述自主避障系统包括导航定位模块、环境感知模块、路径规划模块和驱动模块，所述自主避障装置包括以下步骤：In a second aspect, an embodiment of the present invention further provides an autonomous obstacle avoidance device for a garbage recycling unmanned boat, which is used in an autonomous obstacle avoidance system. The autonomous obstacle avoidance system includes a navigation and positioning module, an environment perception module, a path planning module, and a driving module. The autonomous obstacle avoidance device includes the following steps:
采集单元，用于通过所述导航定位模块采集卫星数据，并通过数据链将其观测值和站点坐标信息一起传送到无人艇上；A collection unit, used to collect satellite data through the navigation and positioning module, and transmit its observation values and site coordinate information to the unmanned boat through a data link;
差分处理单元，用于通过将所述卫星数据和接收到的所述数据链进行实时载波相位差分处理，获得定位结果；A differential processing unit, used for performing real-time carrier phase differential processing on the satellite data and the received data link to obtain a positioning result;
位置获得单元，用于所述环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，获得障碍物和无人艇目标点位的位置信息；A position acquisition unit, used for the environment perception module to obtain the position information of obstacles and unmanned boat target points by fusing the information measured by the millimeter wave radar and the monocular camera;
驱动单元，用于通过所述路径规划模块进行自主决策，通过所述驱动模块驱动所述无人艇达到自动规划路径和自主避障。The driving unit is used to make autonomous decisions through the path planning module, and drive the unmanned boat through the driving module to achieve automatic path planning and autonomous obstacle avoidance.
第三方面，本发明实施例还提供一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现上述实施例任一项的垃圾回收无人艇的自主避障方法中的步骤。In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the autonomous obstacle avoidance method for the garbage recycling unmanned boat of any one of the above embodiments are implemented.
第四方面，本发明实施例还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现上述实施例任意一项所述的垃圾回收无人艇的自主避障方法中的步骤。In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the autonomous obstacle avoidance method for the garbage recycling unmanned boat described in any one of the above embodiments are implemented.
与相关技术相比，本发明通过所述导航定位模块采集卫星数据，并通过数据链将其观测值和站点坐标信息一起传送到无人艇上；通过将所述卫星数据和接收到的所述数据链进行实时载波相位差分处理，获得定位结果；所述环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，获得障碍物和无人艇目标点位的位置信息；通过所述路径规划模块进行自主决策，通过所述驱动模块驱动所述无人艇达到自动规划路径和自主避障。通过融合毫米波雷达的数据和相机获取到的图片信息，可以充分的感知周围环境，为无人艇提供准确高精度的目标信息以及障碍信息，然后通过规划模块实现路径的自主规划；通过所述路径规划模块进行自主决策，从大量经验数据中学习有效的避碰策略，保证在未知的水面环境下依然可以维持避碰策略的稳定性和准确性，同时引入速度障碍法对ddpg算法的训练进行指导，并在失败区域加入大噪声以提高算法训练的效率和准度，路径规划合理、避障方便、准确性高。Compared with the related art, the present invention collects satellite data through the navigation and positioning module, and transmits its observation value and site coordinate information to the unmanned boat through the data link; the positioning result is obtained by performing real-time carrier phase differential processing on the satellite data and the received data link; the environment perception module obtains the position information of obstacles and unmanned boat target points by fusing the information measured by the millimeter wave radar and the monocular camera; the path planning module makes autonomous decisions, and the driving module drives the unmanned boat to achieve automatic path planning and autonomous obstacle avoidance. By fusing the data of the millimeter wave radar and the image information obtained by the camera, the surrounding environment can be fully perceived, and accurate and high-precision target information and obstacle information can be provided for the unmanned boat, and then the planning module realizes the autonomous planning of the path; the path planning module makes autonomous decisions, learns effective collision avoidance strategies from a large amount of empirical data, and ensures that the stability and accuracy of the collision avoidance strategy can still be maintained in an unknown water surface environment. At the same time, the speed obstacle method is introduced to guide the training of the DDPG algorithm, and large noise is added in the failure area to improve the efficiency and accuracy of the algorithm training. The path planning is reasonable, the obstacle avoidance is convenient, and the accuracy is high.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
下面结合附图详细说明本发明。通过结合以下附图所作的详细描述，本发明的上述或其他方面的内容将变得更清楚和更容易理解。附图中：The present invention will be described in detail below in conjunction with the accompanying drawings. The above and other aspects of the present invention will become clearer and easier to understand through the detailed description made in conjunction with the following drawings. In the accompanying drawings:
图1为本发明垃圾回收无人艇的自主避障方法的方法流程图；FIG1 is a flow chart of an autonomous obstacle avoidance method for a garbage recycling unmanned boat according to the present invention;
图2为本发明步骤S2具体的方法流程图；FIG2 is a specific method flow chart of step S2 of the present invention;
图3为本发明步骤S3具体的方法流程图；FIG3 is a specific method flow chart of step S3 of the present invention;
图4为本发明步骤S4具体的方法流程图；FIG4 is a specific method flow chart of step S4 of the present invention;
图5为本发明垃圾回收无人艇的自主避障方法的方法流程图；FIG5 is a flow chart of the autonomous obstacle avoidance method of the garbage recycling unmanned boat of the present invention;
图6为本发明垃圾回收无人艇的自主避障系统框架的框架图；FIG6 is a framework diagram of the autonomous obstacle avoidance system framework of the garbage recycling unmanned boat of the present invention;
图7为本发明多传感器数据融合流程图；FIG7 is a flow chart of multi-sensor data fusion according to the present invention;
图8为本发明速度障碍法辅助训练DDPG算法的框架图；FIG8 is a framework diagram of the speed obstacle method assisted training DDPG algorithm of the present invention;
图9为本发明无人艇运动学模型；FIG9 is a kinematic model of the unmanned boat of the present invention;
图10为本发明障碍物区域与无人艇的夹角示意图；FIG10 is a schematic diagram of the angle between the obstacle area and the unmanned boat of the present invention;
图11为本发明DDPG算法网络流程图；FIG11 is a network flow chart of the DDPG algorithm of the present invention;
图12为本发明速度障碍法辅助训练DDPG算法的框架；FIG12 is a framework of the speed obstacle method assisted training DDPG algorithm of the present invention;
图13为本发明垃圾回收无人艇的自主避障装置的模块图；FIG13 is a block diagram of an autonomous obstacle avoidance device for a garbage recycling unmanned boat according to the present invention;
图14为本发明计算机设备的模块图。FIG. 14 is a module diagram of a computer device of the present invention.
具体实施方式Detailed ways
下面结合附图详细说明本发明的具体实施方式。The specific implementation of the present invention will be described in detail below with reference to the accompanying drawings.
在此记载的具体实施方式/实施例为本发明的特定的具体实施方式，用于说明本发明的构思，均是解释性和示例性的，不应解释为对本发明实施方式及本发明范围的限制。除在此记载的实施例外，本领域技术人员还能够基于本申请权利要求书和说明书所公开的内容采用显而易见的其它技术方案，这些技术方案包括采用对在此记载的实施例的做出任何显而易见的替换和修改的技术方案，都在本发明的保护范围之内。The specific implementation modes/embodiments recorded herein are specific implementation modes of the present invention, which are used to illustrate the concept of the present invention, are explanatory and exemplary, and should not be interpreted as limiting the implementation modes of the present invention and the scope of the present invention. In addition to the embodiments recorded herein, those skilled in the art can also adopt other obvious technical solutions based on the contents disclosed in the claims and the specification of this application, and these technical solutions include any obvious replacement and modification of the embodiments recorded herein, which are within the protection scope of the present invention.
请参考图1-图8所示，其中，图1为本发明垃圾回收无人艇的自主避障方法的方法流程图；图2为本发明步骤S2具体的方法流程图；图3为本发明步骤S3具体的方法流程图；图4为本发明步骤S4具体的方法流程图；图5为本发明垃圾回收无人艇的自主避障方法的方法流程图；图6为本发明垃圾回收无人艇的自主避障系统框架的框架图；图7为本发明多传感器数据融合流程图；图8为本发明速度障碍法辅助训练DDPG算法的框架图。Please refer to Figures 1 to 8, wherein Figure 1 is a method flow chart of the autonomous obstacle avoidance method of the garbage recycling unmanned boat of the present invention; Figure 2 is a specific method flow chart of step S2 of the present invention; Figure 3 is a specific method flow chart of step S3 of the present invention; Figure 4 is a specific method flow chart of step S4 of the present invention; Figure 5 is a method flow chart of the autonomous obstacle avoidance method of the garbage recycling unmanned boat of the present invention; Figure 6 is a framework diagram of the autonomous obstacle avoidance system framework of the garbage recycling unmanned boat of the present invention; Figure 7 is a multi-sensor data fusion flow chart of the present invention; and Figure 8 is a framework diagram of the speed obstacle method assisted training DDPG algorithm of the present invention.
实施例一Embodiment 1
本发明提供一种垃圾回收无人艇的自主避障方法，用于自主避障系统，所述自主避障系统由艇载工控机和岸端上位机组成，艇载工控机和岸端上位机通过通信系统通信连接，岸端上位机用于输入控制量以及数学向上测量参数，其中艇载工控机由四个模块组成，分别为导航定位模块、环境感知模块、路径规划模块和驱动模块。The present invention provides an autonomous obstacle avoidance method for a garbage recycling unmanned boat, which is used for an autonomous obstacle avoidance system. The autonomous obstacle avoidance system consists of a boat-borne industrial control computer and a shore-end host computer. The boat-borne industrial control computer and the shore-end host computer are connected for communication via a communication system. The shore-end host computer is used for inputting control quantities and mathematical upward measurement parameters. The boat-borne industrial control computer consists of four modules, namely a navigation and positioning module, an environment perception module, a path planning module and a driving module.
所述自主避障方法包括以下步骤：The autonomous obstacle avoidance method comprises the following steps:
S1、通过所述导航定位模块采集卫星数据，并通过数据链将其观测值和站点坐标信息一起传送到无人艇上。S1. Collect satellite data through the navigation and positioning module, and transmit its observation values and site coordinate information to the unmanned boat through the data link.
S2、通过将所述卫星数据和接收到的所述数据链进行实时载波相位差分处理，获得定位结果。S2. Obtain a positioning result by performing real-time carrier phase differential processing on the satellite data and the received data link.
S3、所述环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，获得障碍物和无人艇目标点位的位置信息。S3. The environment perception module obtains the location information of obstacles and unmanned boat target points by fusing the information measured by the millimeter wave radar and the monocular camera.
S4、通过所述路径规划模块进行自主决策，通过所述驱动模块驱动所述无人艇达到自动规划路径和自主避障。S4. Make autonomous decisions through the path planning module, and drive the unmanned boat through the driving module to achieve automatic path planning and autonomous obstacle avoidance.
具体的，通过上述S1-S4的方法，导航定位模块通过RTK采集卫星数据，并通过数据链将其观测值和站点坐标信息一起传送给无人艇，然后通过对所采集到的卫星数据和接收到的数据链进行实时载波相位差分处理，得出厘米级的定位结果。环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，确定障碍物和无人艇目标点位的位置，再通过路径规划模块进行自主决策，使无人艇达到自动规划路径和自主避障的功能。通过融合毫米波雷达的数据和相机获取到的图片信息，可以充分的感知周围环境，为无人艇提供准确高精度的目标信息以及障碍信息，然后通过规划模块实现路径的自主规划；通过所述路径规划模块进行自主决策，从大量经验数据中学习有效的避碰策略，保证在未知的水面环境下依然可以维持避碰策略的稳定性和准确性，同时引入速度障碍法对ddpg算法的训练进行指导，并在失败区域加入大噪声以提高算法训练的效率和准度，路径规划合理、避障方便、准确性高。Specifically, through the above-mentioned S1-S4 methods, the navigation and positioning module collects satellite data through RTK, and transmits its observation value and site coordinate information to the unmanned boat through the data link, and then obtains the centimeter-level positioning result by performing real-time carrier phase differential processing on the collected satellite data and the received data link. The environmental perception module determines the position of obstacles and unmanned boat target points by fusing the information measured by the millimeter-wave radar and the monocular camera, and then makes autonomous decisions through the path planning module, so that the unmanned boat can achieve the functions of automatic path planning and autonomous obstacle avoidance. By fusing the data of the millimeter-wave radar and the image information obtained by the camera, the surrounding environment can be fully perceived, and accurate and high-precision target information and obstacle information can be provided to the unmanned boat, and then the planning module can realize autonomous path planning; autonomous decision-making is made through the path planning module, and effective collision avoidance strategies are learned from a large amount of empirical data to ensure that the stability and accuracy of the collision avoidance strategy can still be maintained in an unknown water surface environment. At the same time, the speed obstacle method is introduced to guide the training of the ddpg algorithm, and large noise is added to the failure area to improve the efficiency and accuracy of the algorithm training. The path planning is reasonable, obstacle avoidance is convenient, and the accuracy is high.
在本实施例中，步骤S2具体包括以下子步骤：In this embodiment, step S2 specifically includes the following sub-steps:
S21、通过RANSAC算法在迭代过程消除数据中的相差值。S21. Eliminate the difference values in the data in an iterative process through the RANSAC algorithm.
S22、利用PCA法进行平面拟合。S22. Use PCA method to perform plane fitting.
S23、获得厘米级的定位结果。S23. Obtain centimeter-level positioning results.
其中，RANSAC算法是一种采用迭代的方式从一组包含离群的被观测数据中估算出数学模型的方法。RANSAC算法融合了筛选删除不合格数据的思想，因此在许多环境下，对于有部分错误数据的数据样本，能够更加准确的得到辨识结果。Among them, the RANSAC algorithm is a method that uses an iterative method to estimate a mathematical model from a set of observed data containing outliers. The RANSAC algorithm incorporates the idea of screening and deleting unqualified data, so in many environments, for data samples with some erroneous data, it can obtain more accurate identification results.
具体的，通过RANSAC算法在迭代过程消除数据中的粗差值，可以使得测量数据中的异常值大幅减少，再利用PCA法进行平面拟合，以替代常用的最小二乘法拟合平面。获得厘米级的定位结果。因为最小二乘法虽然可以减小观测向量中的误差，但却忽略了系数矩阵中的误差，会导致拟合出的二维平面精度较差，进而影响接下来的路径规划步骤，甚至影响无人艇航行安全。Specifically, the RANSAC algorithm can be used to eliminate the gross error values in the data in the iterative process, which can greatly reduce the outliers in the measurement data, and then the PCA method is used to fit the plane to replace the commonly used least squares method to fit the plane. Centimeter-level positioning results are obtained. Because the least squares method can reduce the error in the observation vector, but ignores the error in the coefficient matrix, it will lead to poor accuracy of the fitted two-dimensional plane, which will affect the subsequent path planning steps and even affect the navigation safety of the unmanned boat.
更进一步地，所述RANSAC算法具体包括以下子步骤：Furthermore, the RANSAC algorithm specifically includes the following sub-steps:
预设艇载传感器检测到的三维坐标数据集为Ω；从所述数据集中选取符合条件的最小样本子集，计算所述最小子集的模型参数作为初始参数；计算所述Ω与所述初始参数之间的差值；将所述差值与预设的阈值相比较，获得对比结果，并根据所述对比结果为判断条件筛选出不符合条件的点并将其删除；再重复以上过程，不断迭代，最终获得一个数学模型参数。The three-dimensional coordinate data set detected by the onboard sensor is preset as Ω; a minimum sample subset that meets the conditions is selected from the data set, and the model parameters of the minimum subset are calculated as initial parameters; the difference between Ω and the initial parameters is calculated; the difference is compared with a preset threshold to obtain a comparison result, and the points that do not meet the conditions are screened out and deleted based on the comparison result as a judgment condition; the above process is repeated, and it is continuously iterated to finally obtain a mathematical model parameter.
具体的，假设艇载传感器检测到的三维坐标数据集为Ω，从中选取符合条件的最小样本子集，计算该最小子集的模型参数作为初始参数，然后计算Ω与初始参数之间的差值，将差值与设定好的阈值相比较，以此为判断条件筛选出不符合条件的点并将其删除。再重复以上过程，不断迭代，最终估算出一个最佳数学模型参数。检测效果好，方便估算最佳数学模型参数，精确度高。Specifically, assuming that the three-dimensional coordinate data set detected by the onboard sensor is Ω, select the minimum sample subset that meets the conditions, calculate the model parameters of the minimum subset as the initial parameters, and then calculate the difference between Ω and the initial parameters, compare the difference with the set threshold, and use this as the judgment condition to filter out points that do not meet the conditions and delete them. Repeat the above process, iterate continuously, and finally estimate an optimal mathematical model parameter. The detection effect is good, it is convenient to estimate the optimal mathematical model parameters, and the accuracy is high.
更进一步地，所述RANSAC算法得出的结果有用的概率为p，所述p的表达式(1)如下：Furthermore, the probability that the result obtained by the RANSAC algorithm is useful is p, and the expression (1) of p is as follows:
p＝1-[1-ωn]K…(1)；p＝1-[1-ωn ]K …(1);
其中，ω为数据在样本点集中的概率，事先不知道，但可以给树以下鲁棒的值。n为模型拟合一次所需的坐标点的个数，K为实际的迭代次数，所述K的表达式(2)如下：Among them, ω is the probability of the data in the sample point set, which is not known in advance, but the following robust value can be given to the tree. n is the number of coordinate points required for model fitting once, and K is the actual number of iterations. The expression (2) of K is as follows:
通过RANSAC算法预处理后得到点集，利用PCA法计算平面的法向量和所有采样点到目标平面的距离d，所述d的表达式(3)如下：After preprocessing with the RANSAC algorithm, the point set is obtained. The PCA method is used to calculate the plane normal vector and the distance d from all sampling points to the target plane. The expression (3) of d is as follows:
其中xi,yi,zi为采样点坐标，N为采样点个数，该xi,yi,zi分别的均值表达式(4)如下：Where xi , yi ,zi are the coordinates of the sampling points, N is the number of sampling points, and the mean expressions (4) of xi , yi ,zi are as follows:
再求出di的标准差δ，所述标准差δ的表达式(5)如下：Then find the standard deviation δ of di . The expression (5) of the standard deviation δ is as follows:
当di<2δ时，保留该点，通过不断迭代得到最佳二维拟合方程；得到无人艇周边环境信息并进行以上处理之后，使用目标检测算法yolov4对水面的垃圾进行识别，再利用KCF算法对目标垃圾进行跟踪。When di <2δ, this point is retained, and the best two-dimensional fitting equation is obtained through continuous iteration. After obtaining the surrounding environment information of the unmanned boat and performing the above processing, the target detection algorithm yolov4 is used to identify the garbage on the water surface, and then the KCF algorithm is used to track the target garbage.
其中，KCF算法全称是Kernelized Correlation Filters，这个算法不论是在跟踪效果还是跟踪速度上都有十分亮眼的表现。该算法主要使用循环矩阵对样本进行采集，使用快速傅里叶变换对算法进行加速计算。Among them, the full name of the KCF algorithm is Kernelized Correlation Filters. This algorithm has very impressive performance in both tracking effect and tracking speed. The algorithm mainly uses circulant matrices to collect samples and uses fast Fourier transform to accelerate the algorithm calculation.
具体的，通过上述的表达式(1)-(5)，可以提高对水面垃圾识别效率，便于控制无人艇对目标垃圾进行实时追踪，垃圾识别精度高，追踪效果好。Specifically, through the above expressions (1)-(5), the efficiency of identifying surface garbage can be improved, and it is convenient to control the unmanned boat to track the target garbage in real time, with high garbage identification accuracy and good tracking effect.
在本实施例中，步骤S3具体包括以下子步骤：In this embodiment, step S3 specifically includes the following sub-steps:
S31、获取毫米波雷达和单目摄像头所获取的环境信息，其中，所述环境信息包括所述毫米雷达的点云数据和所述单目摄像头的图像数据。S31. Acquire environmental information acquired by the millimeter-wave radar and the monocular camera, wherein the environmental information includes point cloud data of the millimeter-wave radar and image data of the monocular camera.
S32、对所述点云数据和所述图像数据进行预设时间同步处理。S32: Perform preset time synchronization processing on the point cloud data and the image data.
S33、对所述点云数据和所述图像数据分别进行滤波和降噪处理。S33, performing filtering and noise reduction processing on the point cloud data and the image data respectively.
S34、对降噪后的所述图像数据进行目标检测。S34: performing target detection on the image data after noise reduction.
S35、将所述点云数据和所述图像数据进行投影变换，再进行数据融合定位，获得所述无人艇周围三维空间下的障碍物运动信息以及所对应的坐标值。S35, projecting the point cloud data and the image data, and then performing data fusion positioning to obtain obstacle motion information and corresponding coordinate values in the three-dimensional space around the unmanned boat.
通过上述步骤S31-步骤S35的方法，在三维坐标系当中，不利于无人艇的路径规划，需要再将三维空间中的障碍物信息转至二维平面，以达到简化计算量，提高路径规划的实时率这一目的。通过融合毫米波雷达的数据和相机获取到的图片信息，可以充分的感知周围环境，为无人艇提供准确高精度的目标信息以及障碍信息，然后通过规划模块实现路径的自主规划。解决了单一传感器获取信息不足而导致路径规划不合理的问题。利用PCA法替代最小二乘法拟合平面，并通过Ransc法对传感器探测到的数据进行预处理，能提高拟合后的二维平面的精度，减小误差。Through the above-mentioned method of step S31-step S35, in the three-dimensional coordinate system, it is not conducive to the path planning of the unmanned boat. It is necessary to transfer the obstacle information in the three-dimensional space to the two-dimensional plane to simplify the calculation amount and improve the real-time rate of path planning. By integrating the data of the millimeter-wave radar and the image information obtained by the camera, the surrounding environment can be fully perceived, and accurate and high-precision target information and obstacle information can be provided for the unmanned boat, and then the autonomous planning of the path can be realized through the planning module. The problem of unreasonable path planning caused by insufficient information obtained by a single sensor is solved. Using the PCA method instead of the least squares method to fit the plane, and pre-processing the data detected by the sensor through the Ransc method can improve the accuracy of the fitted two-dimensional plane and reduce the error.
在本实施例中，步骤S4具体包括以下子步骤：In this embodiment, step S4 specifically includes the following sub-steps:
S41、根据所述自主决策建立预设的无人艇运行学模型。S41. Establishing a preset unmanned boat operation model according to the autonomous decision.
具体的，为了方便分析，将无人艇看作一个质点，使用航向角的角速度控制无人艇的运动过程，如图9所示。Specifically, for the convenience of analysis, the unmanned boat is regarded as a particle, and the angular velocity of the heading angle is used to control the motion process of the unmanned boat, as shown in FIG9 .
无人艇的运动方程可表示为：The motion equation of the unmanned boat can be expressed as:
其中，vu是无人艇在二维平面中的速度，α是无人艇的航向角，ω是无人艇的角速度。Among them, vu is the speed of the unmanned boat in the two-dimensional plane, α is the heading angle of the unmanned boat, and ω is the angular velocity of the unmanned boat.
在水面航行的过程中，无人艇的航向角以及其角速度满足一下性能约束条件：During the surface navigation, the heading angle and angular velocity of the unmanned boat meet the following performance constraints:
S42、根据所述无人艇运行学模型将深度学习神经网络融合进DPG策略中，获得DDPG算法。S42. According to the unmanned boat operation model, the deep learning neural network is integrated into the DPG strategy to obtain the DDPG algorithm.
具体的，DDPG算法是是将深度学习神经网络融合进DPG的策略学习方法。在该算法中，无人艇通过试错的方式对水面不同情况下的最优路径选择进行学习。用这种方法学习的过程耗时较长，效率不高。因此，设计通过速度障碍法对无人艇的避障选择进行优化，以提高学习效率。Specifically, the DDPG algorithm is a strategy learning method that integrates deep learning neural networks into DPG. In this algorithm, the unmanned boat learns the optimal path selection under different conditions on the water surface by trial and error. The learning process using this method is time-consuming and inefficient. Therefore, the design optimizes the obstacle avoidance selection of the unmanned boat through the speed obstacle method to improve learning efficiency.
假设传感器检测到的障碍为圆形，半径为ro，无人艇半径为rs，则可以通过膨胀障碍圆至R＝rs+ro，将无人艇简化为一个质点；如图10所示。Assuming that the obstacle detected by the sensor is a circle with a radius of ro and the radius of the unmanned boat is rs , the unmanned boat can be simplified to a point mass by expanding the obstacle circle to R = rs + ro , as shown in FIG10 .
S43、根据所述无人艇上的艇载传感器采集的艇载数据。S43, collecting onboard data from onboard sensors on the unmanned boat.
S44、根据所述艇载数据采用速度障碍法进行处理。S44, processing the boat-borne data using a speed obstacle method.
S45、根据处理获取一组样本数据。S45. Obtain a set of sample data according to the processing.
S46、将所述样本数据放入经验池中。S46: putting the sample data into an experience pool.
S47、选择策略动作和做出评价。S47. Select strategic actions and make evaluations.
S48、根据所述策略动作和做出评价选择航向角度。S48. Select a heading angle according to the strategic action and the evaluation.
具体的，通过无人艇上的毫米波雷达和摄像头，可以得到无人艇与障碍物之间的相对速度vuoi和无人艇对应的质点到膨胀后的障碍圆的切线与质点到障碍物圆心Oi的夹角αoi。相对速度矢量与和无人艇与障碍物相对位置之间的夹角为αi。Specifically, the relative speed vuoi between the unmanned boat and the obstacle and the angle αoi between the tangent line from the particle corresponding to the unmanned boat to the expanded obstacle circle and the particle to the obstacle center Oi can be obtained through the millimeter wave radar and camera on the unmanned boat. The angle between the relative speed vector and the relative position of the unmanned boat and the obstacle is αi .
当αi≥αoi时，可知无人艇沿当前航行方向航行时，障碍物对无人艇不构成影响；当αi≤αoi时，无人艇沿当前航行方向航行，可能会与障碍物发生相撞，影响无人艇的航行安全。在速度障碍法中，根据αi和αoi的大小关系对无人艇是否要执行避障动作以及避障所需调整的航向角角度大小做出动作。When αi ≥ αoi , it can be seen that when the unmanned boat sails in the current sailing direction, obstacles will not affect the unmanned boat; when αi ≤ αoi , the unmanned boat sails in the current sailing direction and may collide with obstacles, affecting the navigation safety of the unmanned boat. In the speed obstacle method, the unmanned boat is determined whether to perform obstacle avoidance and the heading angle required to avoid obstacles according to the size relationship between αi and αoi .
DDPG算法是Actor-Critic和DQN算法的结合体，与DQN算法只能作用在离散空间不同，DDPG算法作用在连续空间，所以可以通过神经网络输出连续的调整量，适当修改无人艇航向角，达到避障的目的。DDPG网络结构由现实actor网络、目标actor网络、现实critic网络、目标critic网络这四个网络组成。actor网络执行策略动作，其网络权重参数为θ，输入状态为St，输出动作为at；critic网络给出做出动作的评分Q，使得做出最优选择时的Q值最大，其中网络权重参数为ω,输入状态st和动作at，输出为评价值Q。The DDPG algorithm is a combination of the Actor-Critic and DQN algorithms. Unlike the DQN algorithm, which can only work in discrete space, the DDPG algorithm works in continuous space, so it can output continuous adjustments through the neural network, appropriately modify the heading angle of the unmanned boat, and achieve the purpose of obstacle avoidance. The DDPG network structure consists of four networks: the real actor network, the target actor network, the real critic network, and the target critic network. The actor network executes the strategy action, and its network weight parameter is θ, the input state is St, and the output action is at; the critic network gives the score Q of the action, so that the Q value is maximized when the optimal choice is made. The network weight parameter is ω, the input state st and the action at, and the output is the evaluation value Q.
actor网络更新采用梯度下降法，如下式所示：The actor network is updated using the gradient descent method, as shown below:
m为样本数据的采样个数。critic网络通过均方误差损失函数进行参数更新。m is the number of sample data. The critic network updates its parameters through the mean square error loss function.
γ为奖励折扣因子。如图11所示。γ is the reward discount factor, as shown in Figure 11.
其中状态S由艇载传感器获得，输入到现实actor得到动作a，对无人艇施加动作a，无人艇与环境交互返回下一时刻的状态S’和奖励r’，从而得到一组样本数据(S,a,r,S’)，把其放入经验池中。把(S,a,r,S’)中的S和a输入到现实Critic中，得到现实Q(S,a)值,令Q＝Q(S,a)。然后把(S,a,r,S’)中的S’输入到目标Actor中，得到动作a’。并把S’和a’一起输入到目标Critic中，得到Q(S’，a’)于是目标Q值为Q’＝r+γ×Q(S’，a’)值。再把Q’看成标签，更新现实critic使得输出Q尽量接近标签Q’，再更新现实Actor，因为现实Actor输出的动作，在现实critic里给出了Q值，更新现实Actor使得Q值输出最大，循环更新采样，以达到贴近最优选择的操作。The state S is obtained by the onboard sensor, and is input into the real actor to obtain action a. Action a is applied to the unmanned boat. The unmanned boat interacts with the environment and returns the state S' and reward r' of the next moment, thereby obtaining a set of sample data (S, a, r, S'), which is put into the experience pool. Input S and a in (S, a, r, S') into the real critic to obtain the real Q(S, a) value, and let Q = Q(S, a). Then input S' in (S, a, r, S') into the target actor to obtain action a'. And input S' and a' together into the target critic to obtain Q(S', a') so that the target Q value is Q' = r + γ × Q(S', a'). Then regard Q' as a label, update the real critic so that the output Q is as close to the label Q' as possible, and then update the real actor, because the action output by the real actor gives a Q value in the real critic. Update the real actor to maximize the Q value output, and update the sampling in a loop to achieve an operation close to the optimal choice.
通过速度障碍法辅助训练DDPG算法的框架如图12所示。在DDPG算法进行避障训练时，会在目标actor网络输出动作时添加高斯白噪声将该网络执行的确定值动作变为随机值动作，以提高DDPG算法的探索能力。The framework of the DDPG algorithm assisted in training by the speed obstacle method is shown in Figure 12. When the DDPG algorithm is performing obstacle avoidance training, Gaussian white noise is added when the target actor network outputs an action to convert the determined value action executed by the network into a random value action, so as to improve the exploration ability of the DDPG algorithm.
at’＝at+EN…(10)；at ′＝at +EN…(10);
式中：at为原actor网络输出动作；EN为符合高斯分布的随机探索因子；at’为加入探索因子后具备随机探索能力的输出动作。采用的探索噪声如下式：Where: at is the output action of the original actor network; EN is the random exploration factor that conforms to the Gaussian distribution; at ' is the output action with random exploration ability after adding the exploration factor. The exploration noise used is as follows:
EN～N(μ＝0，δ＝0.5)…(11)。EN～N(μ=0, δ=0.5)…(11).
更优的，所述自主避障方法还包括以下子步骤：Preferably, the autonomous obstacle avoidance method further comprises the following sub-steps:
S5、所述DDPG算法做出错误选择。S5. The DDPG algorithm makes an incorrect choice.
S6、将所述DDPG算法退回为上个状态。S6. Return the DDPG algorithm to the previous state.
S7、DDPG网络输出动作值并提高噪声。S7, DDPG network outputs action value and increases noise.
S8、根据所述动作值控制所述无人艇执行相应动作，并得到样本数据。S8. Control the unmanned boat to perform corresponding actions according to the action value, and obtain sample data.
S9、根据所述样本数据进行DDPG网络更新。S9. Update the DDPG network according to the sample data.
具体的，在许多实验中，DDPG算法会在探索失败区域反复触发训练失败条件，学习效率较差。本设计通过在失败区域提高探索随机性，以提高算法的学习效率。提高方法如下式所示：Specifically, in many experiments, the DDPG algorithm repeatedly triggers training failure conditions in the exploration failure area, resulting in poor learning efficiency. This design improves the learning efficiency of the algorithm by increasing the randomness of exploration in the failure area. The improvement method is shown in the following formula:
EN～0.5(N(-1，0.5)+N(1，0.5))…(12)；EN～0.5(N(-1,0.5)+N(1,0.5))…(12);
与算法中的原噪声相比，在失败区域将探索幅度从[-0.5,0.5]之间向两侧偏移至[-1.5,0.5]和[0.5,1.5]的区间，提高了随机探索的幅度，以提高找到正确航行方向的速度。在跳出失败区域后，再继续对失败区域进行反复多次的训练，以积累大量的样本数据，提高算法选择航向角的准确率。Compared with the original noise in the algorithm, the exploration amplitude in the failure area is shifted from [-0.5, 0.5] to [-1.5, 0.5] and [0.5, 1.5] on both sides, increasing the amplitude of random exploration to increase the speed of finding the correct navigation direction. After jumping out of the failure area, the failure area is repeatedly trained to accumulate a large amount of sample data and improve the accuracy of the algorithm in selecting the heading angle.
具体的，通过利用DDPG算法的多维度特征提取能力，从大量经验数据中学习有效的避碰策略，保证在未知的水面环境下依然可以维持避碰策略的稳定性和准确性，同时引入速度障碍法对DDPG算法的训练进行指导，并在失败区域加入大噪声以提高算法训练的效率和准度，有效的克服了DDPG算法训练收敛速度慢，样本数据利用率低的问题。Specifically, by utilizing the multi-dimensional feature extraction capability of the DDPG algorithm, effective collision avoidance strategies are learned from a large amount of empirical data to ensure that the stability and accuracy of the collision avoidance strategy can be maintained in unknown water surface environments. At the same time, the speed barrier method is introduced to guide the training of the DDPG algorithm, and large noise is added to the failure area to improve the efficiency and accuracy of the algorithm training, effectively overcoming the problems of slow convergence speed of DDPG algorithm training and low sample data utilization.
实施例二Embodiment 2
请参阅附图13所示，图13为本发明垃圾回收无人艇的自主避障装置的模块图。本发明实施例还提供一种垃圾回收无人艇的自主避障装置200，用于自主避障系统，所述自主避障系统包括导航定位模块、环境感知模块、路径规划模块和驱动模块，所述自主避障装置200包括以下步骤：Please refer to Figure 13, which is a module diagram of the autonomous obstacle avoidance device of the garbage recycling unmanned boat of the present invention. The embodiment of the present invention also provides an autonomous obstacle avoidance device 200 for the garbage recycling unmanned boat, which is used in an autonomous obstacle avoidance system. The autonomous obstacle avoidance system includes a navigation and positioning module, an environment perception module, a path planning module and a driving module. The autonomous obstacle avoidance device 200 includes the following steps:
采集单元201，用于通过所述导航定位模块采集卫星数据，并通过数据链将其观测值和站点坐标信息一起传送到无人艇上；The acquisition unit 201 is used to collect satellite data through the navigation and positioning module, and transmit its observation values and site coordinate information to the unmanned boat through the data link;
差分处理单元202，用于通过将所述卫星数据和接收到的所述数据链进行实时载波相位差分处理，获得定位结果；The differential processing unit 202 is used to obtain a positioning result by performing real-time carrier phase differential processing on the satellite data and the received data link;
位置获得单元203，用于所述环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，获得障碍物和无人艇目标点位的位置信息；The position acquisition unit 203 is used for the environment perception module to obtain the position information of obstacles and unmanned boat target points by fusing the information measured by the millimeter wave radar and the monocular camera;
驱动单元204，用于通过所述路径规划模块进行自主决策，通过所述驱动模块驱动所述无人艇达到自动规划路径和自主避障。The driving unit 204 is used to make autonomous decisions through the path planning module, and drive the unmanned boat through the driving module to achieve automatic path planning and autonomous obstacle avoidance.
具体的，通过采集单元201用于通过所述导航定位模块采集卫星数据，并通过数据链将其观测值和站点坐标信息一起传送到无人艇上；差分处理单元202用于通过将所述卫星数据和接收到的所述数据链进行实时载波相位差分处理，获得定位结果；位置获得单元203用于所述环境感知模块通过融合毫米波雷达与单目摄像头所测得的信息，获得障碍物和无人艇目标点位的位置信息；驱动单元204用于通过所述路径规划模块进行自主决策，通过所述驱动模块驱动所述无人艇达到自动规划路径和自主避障。通过融合毫米波雷达的数据和相机获取到的图片信息，可以充分的感知周围环境，为无人艇提供准确高精度的目标信息以及障碍信息，然后通过规划模块实现路径的自主规划；通过所述路径规划模块进行自主决策，从大量经验数据中学习有效的避碰策略，保证在未知的水面环境下依然可以维持避碰策略的稳定性和准确性，同时引入速度障碍法对ddpg算法的训练进行指导，并在失败区域加入大噪声以提高算法训练的效率和准度，路径规划合理、避障方便、准确性高。Specifically, the acquisition unit 201 is used to collect satellite data through the navigation and positioning module, and transmit its observation values and site coordinate information to the unmanned boat through the data link; the differential processing unit 202 is used to obtain the positioning result by performing real-time carrier phase differential processing on the satellite data and the received data link; the position acquisition unit 203 is used for the environmental perception module to obtain the position information of obstacles and unmanned boat target points by fusing the information measured by the millimeter wave radar and the monocular camera; the driving unit 204 is used to make autonomous decisions through the path planning module, and drive the unmanned boat through the driving module to achieve automatic path planning and autonomous obstacle avoidance. By integrating the data from the millimeter-wave radar and the image information obtained by the camera, the surrounding environment can be fully perceived, and accurate and high-precision target information and obstacle information can be provided to the unmanned boat. Then, the planning module can realize autonomous path planning. The path planning module makes autonomous decisions and learns effective collision avoidance strategies from a large amount of empirical data to ensure that the stability and accuracy of the collision avoidance strategies can be maintained in unknown water environments. At the same time, the speed obstacle method is introduced to guide the training of the DPG algorithm, and large noise is added to the failure area to improve the efficiency and accuracy of the algorithm training. The path planning is reasonable, obstacle avoidance is convenient, and accuracy is high.
实施例三Embodiment 3
请参考图14所示，图14为本发明计算机设备的模块图。本发明实施例还提供一种计算机设备，包括存储器301、处理器302及存储在所述存储器301上并可在所述处理器302上运行的计算机程序，所述处理器302执行所述计算机程序时实现上述实施例一的垃圾回收无人艇的自主避障方法中的步骤。Please refer to Figure 14, which is a module diagram of the computer device of the present invention. The embodiment of the present invention also provides a computer device, including a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302, and the processor 302 implements the steps of the autonomous obstacle avoidance method of the garbage collection unmanned boat in the first embodiment when executing the computer program.
实施例四Embodiment 4
本发明实施例还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现上述实施例一的垃圾回收无人艇的自主避障方法中的步骤。An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the autonomous obstacle avoidance method for the garbage recycling unmanned boat in the above-mentioned embodiment 1 are implemented.
以上所述仅为本发明的优选实施例而已，并不用于限制本发明，对于本领域的技术人员来说，本发明可以有各种更改和变化。凡在本发明的精神和原则之内，所作的任何纂改、等同替换、改进等，均应包含在本发明的权利要求范围之内。The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210831313.6ACN115373383B (en) | 2022-07-15 | 2022-07-15 | Autonomous obstacle avoidance method and device for garbage recycling unmanned ship and related equipment |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210831313.6ACN115373383B (en) | 2022-07-15 | 2022-07-15 | Autonomous obstacle avoidance method and device for garbage recycling unmanned ship and related equipment |
| Publication Number | Publication Date |
|---|---|
| CN115373383A CN115373383A (en) | 2022-11-22 |
| CN115373383Btrue CN115373383B (en) | 2024-07-19 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210831313.6AActiveCN115373383B (en) | 2022-07-15 | 2022-07-15 | Autonomous obstacle avoidance method and device for garbage recycling unmanned ship and related equipment |
| Country | Link |
|---|---|
| CN (1) | CN115373383B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116243720B (en)* | 2023-04-25 | 2023-08-22 | 广东工业大学 | A 5G network-based AUV underwater object-finding method and system |
| CN116499467B (en)* | 2023-06-07 | 2025-09-23 | 广东利元亨智能装备股份有限公司 | A robot mobile obstacle avoidance method, device, equipment and medium |
| CN119665975B (en)* | 2024-12-09 | 2025-09-23 | 北京理工大学 | A method and system for seamless indoor and outdoor switching navigation of unmanned aerial vehicles |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109298708A (en)* | 2018-08-31 | 2019-02-01 | 中船重工鹏力（南京）大气海洋信息系统有限公司 | A kind of unmanned boat automatic obstacle avoiding method merging radar and photoelectric information |
| CN113917454A (en)* | 2021-10-11 | 2022-01-11 | 上海大学 | A method and system for fusion detection of unmanned boat surface targets |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112560972B (en)* | 2020-12-21 | 2021-10-08 | 北京航空航天大学 | Target detection method based on millimeter wave radar prior location and visual feature fusion |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109298708A (en)* | 2018-08-31 | 2019-02-01 | 中船重工鹏力（南京）大气海洋信息系统有限公司 | A kind of unmanned boat automatic obstacle avoiding method merging radar and photoelectric information |
| CN113917454A (en)* | 2021-10-11 | 2022-01-11 | 上海大学 | A method and system for fusion detection of unmanned boat surface targets |
| Publication number | Publication date |
|---|---|
| CN115373383A (en) | 2022-11-22 |
| Publication | Publication Date | Title |
|---|---|---|
| CN115373383B (en) | Autonomous obstacle avoidance method and device for garbage recycling unmanned ship and related equipment | |
| EP3405845B1 (en) | Object-focused active three-dimensional reconstruction | |
| CN103278170B (en) | Based on mobile robot's cascade map creating method that remarkable scene point detects | |
| Sevastopoulos et al. | A survey of traversability estimation for mobile robots | |
| Chen et al. | Robot navigation with map-based deep reinforcement learning | |
| Begum et al. | Integrated fuzzy logic and genetic algorithmic approach for simultaneous localization and mapping of mobile robots | |
| Saleem et al. | Steering angle prediction techniques for autonomous ground vehicles: a review | |
| CN112762957A (en) | Multi-sensor fusion-based environment modeling and path planning method | |
| CN119469168B (en) | Autonomous navigation method and system of quadruped robot for special environments | |
| CN116576857A (en) | Multi-obstacle prediction navigation obstacle avoidance method based on single-line laser radar | |
| CN114998276B (en) | A real-time detection method for robot dynamic obstacles based on 3D point cloud | |
| Min et al. | Autonomous driving in unstructured environments: How far have we come? | |
| CN118502426A (en) | Mobile robot based on dynamic control obstacle function and self-adaptive optimal control | |
| CN119910648B (en) | Bionic robot control method and system based on AI learning | |
| CN117970928A (en) | Path planning method for multiple mobile robots based on ROS system | |
| KR20230036243A (en) | Real-time 3D object detection and tracking system using visual and LiDAR | |
| Short et al. | A bio-inspired algorithm in image-based path planning and localization using visual features and maps | |
| CN118752498A (en) | Autonomous navigation method and system for a steel bar tying robot | |
| CN117589167A (en) | Unmanned aerial vehicle routing inspection route planning method based on three-dimensional point cloud model | |
| CN119310859A (en) | A dynamic target tracking method and system based on image feedback | |
| CN119902222A (en) | BIM and measurement robot collaborative intelligent measurement method, system, equipment and medium | |
| Masmoudi et al. | Autonomous car-following approach based on real-time video frames processing | |
| CN119247957A (en) | A mobile robot capable of autonomously building a map in an unknown closed space and a working method thereof | |
| Yildiz et al. | CNN based sensor fusion method for real-time autonomous robotics systems | |
| CN117470241A (en) | An adaptive mapless navigation method and system for uneven terrain |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |