Benchmarking Actor-Critic Deep Reinforcement Learning Algorithms for Robotics Control with Action Constraints

This study presents a benchmark for evaluating action-constrained reinforcement learning (RL) algorithms. In RL, each action taken by the system must comply with certain constraints. These constraints are crucial ensuring feasibility and safety of actions in real-world systems. We evaluate existing algorithms their novel variants across multiple robotics control environments, encompassing constraint types. Our evaluation provides first in-depth perspective field, revealing surprising insights, including effectiveness straightforward baseline approach. The problems associated code utilized our experiments made available online at github.com/omron-sinicx/action-constrained-RL-benchmark further research development.