A Constrained Workspace Entrance Crossing Motion Generation Scheme Synthesized by Recurrent Neural Networks for Redundant Robot Manipulators

In order to enable redundant robot manipulators to safely cross constrained workspace entrances and complete end-effector tasks, a novel recurrent neural network based constrained workspace entrance crossing motion generation (CECMG) scheme is proposed and exploited. Specifically, the constrained workspace entrance safety crossing condition is modeled as an inequality type remote center of motion (RCM) constraint. By constraining the critical point of the manipulator within the predefined range of the RCM, the proposed scheme ensures that the redundant manipulator safely passes through the constrained workspace entrance. Then, a linear variational inequality-based primal-dual neural network is utilized to dynamically solve the proposed scheme and generate the motion sequence for the manipulator in real-time. Two validation experiments confirmed that the proposed CECMG scheme can be used for a robotic arm to cross into a constrained workspace entrance from different directions and complete end-effector tasks. In addition, comparative experiments validate that the proposed CECMG scheme addresses the limitations of the existing equation type RCM method that must be artificially placed in order to work behind the constrained workspace entrance. Physical experiments are conducted to verify the effectiveness and implementability of the proposed CECMG scheme.