Safe deployment of a reinforcement learning robot using self stabilization

• DOI: 10.1016/j.iswa.2022.200105
• Publication Date: 2022-07-26T17:54:40Z
• Authors: Nanda Kishore Sre...; Shrisha Rao
• Keywords: Self stabilization; 0209 industrial biotechnology; Safety in robotics; Electronic computers. Computer science; Reinforcement learning; Q300-390; QA75.5-76.95; 02 engineering and technology; Cybernetics

In toy environments like video games, a reinforcement learning agent is deployed and operates within the same state space in which it was trained. However, robotics applications such as industrial systems or autonomous vehicles, this cannot be guaranteed. A robot can pushed out of its training by some unforeseen perturbation, may cause to go into an unknown from has not been trained move towards goal. While most prior work area RL safety focuses on ensuring phase, paper safe deployment that already operate space. This defines condition action spaces, if satisfied, guarantees robot’s recovery independently. We also propose strategy design facilitate finite number steps after perturbation. implemented tested against standard model, results indicate significant improvement performance.