Research On Robot Manipulation Planning Based On Segmentation And Learning Of Hand Pose Sequence

The wide application of robots in production and life can not only help humans with heavy,repetitive and dangerous labor,but also greatly improve the qua lity and efficiency of task operations.Currently,the improvement of robots' autonomy has become an important trend.However,the autonomous manipulation planning of robots under the constraints of complex tasks is still a difficult problem.The existing manipulation planning methods are facing challenges and urgently need to break through the technical bottleneck.Firstly,because task constraints are implicitly defined and difficult to obtain directly,this paper proposes a method for segmentation and le arning of hand pose sequence based on two-stage peak detection.In order to obtain hand pose sequences from visual teaching,a hand pose estimation method is proposed to collect hand pose sequences.To get subsequences containing task constraints,segmenta tion of hand pose sequence based on two-stage peak detection is proposed.A hand pose sequence learning method for manipulation planning under task constraints is introduced to obtain the parametric description of task constraints.The feasibility of the proposed segmentation method is verified through experiments.Secondly,robots complete tasks by executing paths generated by manipulation planning.However,under task constraints,existing manipulation planning methods cannot achieve feasible results that balance path quality,planning efficiency and success rate.A method for manipulation planning under task constraints is proposed.To represent task constraints implicitly defined in configuration space,a method for building approximation model of task constraints based on metric learning is proposed.To conduct manipulation planning under task constraints with high path quality,planning efficiency and success rate,this thesis proposes a method for robot manipulation planning based on guidance of task constraint approximation model.Finally,the feasibility of robot manipulation planning based on segmentation and learning of hand pose sequence is verified.Key processes in the proposed method are tested through experiments.The experiments include experi ment on segmentation and learning of hand pose sequence,which evaluates segmented subsequence's coverage of ground truth subsequence;robot manipulation planning experiment,which collects and compares data of path quality,planning efficiency and success rate with existing methods.