Bridging Locomotion and Manipulation Using Reconfigurable Robotic Limbs via Reinforcement Learning

Locomotion and manipulation are two essential skills in robotics but often divided or decoupled into separate problems. It is widely accepted that the topological duality between multi-legged locomotion multi-fingered shares an intrinsic model. However, a lack of research remains to identify data-driven evidence for further research. This paper explores unified formulation loco-manipulation problem using reinforcement learning (RL) by reconfiguring robotic limbs with overconstrained design robots. Such reconfiguration allows adopting co-training architecture towards policy. As result, we find support transferability single RL policy multilayer perceptron graph neural network. We also demonstrate Sim2Real transfer learned prototype. work expands knowledge frontiers on learning-based applied novel platform limbs.