Hybrid OSC-RL Control for Task Optimization of Dual-Arm Robots

In this work we present a strategy to solve the task optimization problem for dual-arm mobile manipulators in the context of agricultural tasks. The strategy combines a Reinforcement Learning (RL) agent with a low-level Operational Space Controller (OSC). The agent is responsible for motion planning, as well as compensatory torque computation. Preliminary results obtained through physically accurate simulation using MuJoCo show that the method proposed achieves a higher task success rate in task completion.