Task‐space control is an important job in robotics and robot control, where the robot end effector is forced to reach the desired reference position and orientation in the workspace. Reinforcement learning (RL) has been used for robot control based on model‐free and online algorithms. RL methods for robot control include three subclasses: value iteration, policy iteration, and policy search. This chapter discusses the classical joint and task space control schemes, which give the theory and the first solution of the inverse and velocity kinematics problems. It provides the RL approach, which is inspired in the classical solutions. RL for robot control can be regarded as a Markov decision process. The chapter compares multi‐agent RL with the following classical methods: the Jacobian‐based augmented method, the Jacobian‐based singularity avoidance, the Jacobian‐based singular‐value‐decomposition, and the multilayer neural networks. The curse of dimensionality in RL and controllability problems for redundant robot control are overcome.
Redundant Robots Control Using Multi‐Agent Reinforcement Learning
2021-10-05
19 pages
Article/Chapter (Book)
Electronic Resource
English
Springer Verlag | 2025
|