Research On Motion Planningand Path Optimization For Manipulator Based On Gaussian Mixture Models

Traditional methods including manual teaching and offline programming,which are used to control the action of the manipulator.However these methods have many shortcomings,such as cumbersome process,high cost and low efficiency,which can no longer meet the requirements of industrial manufacturing field nowadays.In order to improve the intelligence and practicability of robot,and ensure the safety and stability of manipulators,motion planning of the robot and path optimization have become the focus of research in recent years.The process of autonomous obstacle avoidance is called the motion planning of manipulator in configuration space.The sampling-based motion planning method uses collision detection to obtain obstacle information which doesn't require the deterministic model and the description of the environment.Therefore,this method can effectively explore the connectivity of high-dimensional configuration space,it's also an important method to solve the problem of manipulator motion planning.However,in complex planning scenarios,not only collision avoidance constraints should be satisfied,but also computational efficiency and path quality should be ensured.Therefore,this paper proposes an adaptive sampling strategy and collision checker combined with Gaussian Mixture Model(GMM)to improve the operational efficiency of sample-based motion planning algorithm.Then the path asymptotic optimization algorithm is used to eliminate redundant actions and converge to a short path.The path is used as the initial value of the stochastic path smooth algorithm.After repeated iterations,a continuous smooth path which satisfies both collision avoidance and joint constraints is obtained.The algorithm is verified by simulation and experiment.The main research contents are listed as follows:Firstly,the problems about forward and inverse kinematics of the UR5 robot are solved.The sample-based motion planning algorithm is analyzed and its collision detection module which occupies the longest computational time is the bottleneck of computation efficiency.On the one hand,an adaptive sampling strategy is proposed.Firstly,in different scenes,robots' controller use Gaussian method with adjustable variance to acquire the sample set which is used to train the GMM fitting the target region.Because GMM can guide sampling,the planning time is shortened by narrowing the sampling range.On the other hand,due to the complexity of traditional collision detection method such as AABB bounding box,this paper trains the GMM fitting high-dimensional obstacle space.By calculating the probability of the sample of this model,whether the robot collides or not can be quickly judged.The model training method is also improved to reduce the collision prediction error.The effectiveness of the above methods and the improvement of the calculation efficiency of the GMM based on motion planning algorithm are both verified by experiments.Secondly,the path optimization method is studied in this paper to solve the poor path obtained by above motion planning algorithm.On the one hand,unnecessary actions in the path can be eliminated and chatter can be reduced;On the other hand,a path hybrid algorithm is designed to shorten path length.By generating a new path that combines the optimal segments of multiple paths,the problem of poor consistency and non-optimal path of each planning is solved.The deficiency of randomness of the algorithm is made up too.An interactive path asymptotic optimal algorithm is designed which combines their advantages,and the above two methods alternately execute from micro and macro aspects to generate an asymptotically optimal path.On this basis,a stochastic trajectory smooth algorithm is proposed,which relies on the noise trajectories to explore the low-cost smooth path.The cost function is designed for collision avoidance and smoothness so as to ensure the safe and stable motion of the robot.The effectiveness of the above methods and the improvement of path quality after optimization are verified by experiments.Finally,the robot simulation and experimental platform are set based on ROS system.First,the simulation and experiment platform are built,and RealSense depth camera is used to obtain image information.The above motion planning algorithm and path optimization method are also verified by practical experiments: The sampling strategy and collision detection methods based on GMM are applied in the sample-based motion planning algorithm with the UR5 robot for different motion planning tasks,thus the improvement of computational efficiency about the GMM based motion planning algorithm and the improvement of path quality by optimization algorithm in different motion planning tasks are further verified.