High-Dimensional Inverse Kinematics and Self-Reconfiguration Kinematic Control

This paper addresses two unique challenges for self-reconfigurable robots to perform dexterous locomotion and manipulation in difficult environments: highdimensional inverse kinematics (HDIK) for > 100 degrees of freedom, and selfreconfiguration kinematic control (SRKC) where the workspace targets at which connectors are to meet for docking are not known a priori. These challenges go beyond the state-of-the-art because traditional manipulation techniques (e.g., Jacobian-based) may not be stable or scalable, and alternative approaches (e.g., genetic algorithms or neural networks) provide no guarantees of optimality or convergence. This paper proposes a new technique called Provably-convergent Swarm-based Inverse Kinematics (PSIK) that extends Branch and Bound Particle Swarm Optimization with a unique approach for dynamic target adaptation for selfreconfiguration. The PSIK algorithm can find globally optimal solutions for both HDIK and SRKC to any precision requirement (i.e., positive error tolerance) in finite or real-time for tree structures of self-reconfigurable robots. This algorithm is implemented and validated in high-fidelity, physics-based simulation using SuperBot as prototype modules. The results are very encouraging and provide feasible solutions for dextrous locomotion, manipulation, and self-reconfiguration.