| Autonomous Surface Vessels(ASVs),known as Unmanned Surface Vessels(USVs),are one of the hotspots of military and civilian unmanned technology competition at home and abroad.Compared with the military field,the development of ASVs in the civilian field is relatively lagging behind.Therefore,it has been widely concerned by researchers.Unlike the ASVs in the military field,the requirements in the civilian field are more targeted such as intelligent security of lakes in the city,water surface search and rescue,water quality online monitoring,water surface garbage cleaning,and so on.For the aforementioned needs,it is particularly important to design an ASV capable of completing tasks autonomously.Based on the research of laboratory teachers' mechanical structure,this dissertation mainly completes the hardware platform design of ASVs control system,the design of water surface obstacle detection algorithm based on lidar and the design of path planning algorithm based on improved Q-Learning algorithm.In the control system hardware platform,considering the special safety requirements of water surface navigation,the system control adopts dual-line redundant backup,which greatly improves the reliability of ASVs control.Compared with the marine environment,the inland lake channel environment in the civil field is more complex,with more interference targets and higher requirements for autonomous cruise of ASVs.Therefore,it poses a challenge to the intelligent perception and target recognition of water surface environment.At present,the target perception and recognition of water surface environment at home and abroad mainly use millimeter-wave radar technology,which has the problems of poor anti-interference ability and low resolution.Lidar has the advantages of high resolution,strong anti-interference ability and small size,which can make up for the defects of millimeter-wave radar,but there is not much research on the image processing algorithm of lidar in the water surface environment and the obstacle avoidance algorithm of the combination of lidar and ASVs.First,there is the problem that the lidar image processing is not close to the actual navigation environment,and the problems of isolated points on the water surface and safe navigation distance are not considered.Second,the obstacle avoidance algorithm and the lidar data are not close enough,and dynamic obstacle avoidance is difficult to achieve.Therefore,this project focuses on completing the following work:First,in terms of lidar image processing,preprocessing and filtering of lidar data(image morphology processing),and clustering analysis of obstacle environments based on distance and reflectivity(K nearest neighbor clustering),mainly eliminate isolated points caused by water surface fluctuation and sunlight reflection,and complete obstacle detection and recognition;Second,in the path planning and obstacle avoidance parts,aiming at all kinds of flaws of the classical Q-Learning algorithm,this paper proposes anew rewards and punishment function,improved the classcial Q-Learning algorithm easy to fall into local optimum of faults.By integrating the navigation and obstacle avoidance behaviors of the ASV during navigation,it is more in line with the actual navigation environment of the ASV.Dynamic planning and dynamic obstacle avoidance are realized by constantly updating the iterative value function(Q(s,a))to predict the trajectory of the ASVs.Finally,in order to verify the feasibility of the designed algorithm,we conducted computer simulation experiments on the algorithm,as well as actual navigation tests on the water surface of the reservoir and Suzhou River.The comparison of path planning in simulation experiments shows the effectiveness of the improved algorithm.In the actual navigation test,the ASV can complete the requirements of autonomous cruise and dynamic obstacle avoidance according to the changes of different environments,thus the feasibility and effectiveness of the entire system are verified. |
PDF Full Text Request