Research On Obstacle Avoidance Algorithms Of Robotic Arm Based On Dynamical Movement Primitives

In order to replace humans to complete complex production tasks,robots are often required to be equipped with high adaptability and reliability at the level of movement planning.The traditional movement planning methods suffer from disadvantages like slow convergence or easy to fall into the local minimum state.In contrast,learning from demonstration(Lf D)methods endow the robot humanoid characteristics,which can intuitively and conveniently allow the robot to imitate and replicate.The trajectories obtained by the current demonstration meet the requirements of tasks in an efficient and reasonable way,and meanwhile,it can be generalized on the basis of the original demonstrated trajectory,thereby adapting to the new environment.However,the environment in real task is not always static.How to make the robot react to a certain obstacle in an unstructured environment while maintaining the advantages of the Lf D methods has been a hot issue in the research of trajectory planning in recent years.Based on Dynamical Movement Primitives(DMPs),an existing typical Lf D method,this paper combines the information of obstacles in different states in the environment,and proposes a method to effectively avoid the influence of obstacles according to local conditions.Firstly,an obstacle avoidance method is proposed based on the geometric information of static obstacles.According to the shape characteristics of obstacles,a unified description is adopted,combined with DMPs to automatically adjust the position of the trajectory during the generalization process,so as to achieve successful obstacle avoidance.We analyzed the effect of the obstacle avoidance method on the maintenance and inheritance of the Lf D characteristics.Besides,we optimized the method employing transition curves generated by quintic spline interpolation for the local unsmooth problem,and compared it with one of other smoothing algorithms to verify its feasibility.Secondly,the obstacle avoidance method based on the geometric information of moving obstacles and the method based on the combination of DMPs and improved DWA are respectively proposed for the environment where an obstacle is moving.The former one uses a rectangular hazard zone to uniformly describe the movement of obstacles,generates avoidance trajectories according to a certain bias principle,and uses PD tracking to converge to the target point.However,the latter one adapts to movement by improving the Dynamic Window Algorithm(DWA)and integrate it into the Lf D framework.The avoidance effects of the two methods are analyzed and compared,and the specific environment they respectively adapt to is discussed.Finally,a simulation was carried out for the verification of the algorithms,and a physical Lf D platform was built,while different task models were set up to make the robot generalize and execute.The algorithms were verified in the Pybullet simulation engine,and then the simulation communicated with the real manipulator to ensure the safety of the execution.The trajectories of pouring water,planar writing,and spatial letter G were selected,and the scene where obstacles were in different states was set up for experimental verification,which proved the validity and feasibility of the obstacle avoidance algorithms in different situation.