Generating of Task-Based Controls for Joint-Arm Robots with Simulation-based Reinforcement Learning

The paper investigates how a robot control for a pick-and-place application can be learned by simulation using the Q-Learning method, a special Reinforcement Learning approach. Furthermore, a post-optimization approach to improve a learned strategy is presented. Finally, it is shown how the post-optimized strategy can be automatically transformed into an executable control using the simulation-based control approach.