Abstract—Ensuring a high level of safety is essential for collision avoidance in real-world robotic applications. Traditional Reinforcement Learning (RL)-based collision avoidance methods offer adaptability but lack safety guarantees, especially in uncertain and dynamic environments. To address this, we propose a novel safe reinforcement learning (SafeRL) framework called Control Recovery and Barrier Function (CRBF), which enhances safety by sequentially applying different control strategies based on the robot’s proximity to obstacles. The CRBF categorizes risk into three distinct levels and adaptively switches between a vanilla RL-based policy, Control Barrier Function (CBF), and a Recovery Function (RF) to prevent collisions and recover from critical situations. In addition, we introduce a constraint-aware training strategy that incorporates these sequential safety mechanisms during policy updates. We validate our method in both simulated and real-world environments, where CRBF outperforms conventional methods, with improvements of up to 22.5% in collision avoidance success rates, particularly in challenging dynamic scenarios.

Keywords—collision avoidance, safe reinforcement learning, control barrier functions, recovery function

Cite: Gyuyong Hwang and Gyuho Eoh, "Safe Reinforcement Learning Using Sequential Constraints for Collision Avoidance," International Journal of Mechanical Engineering and Robotics Research, Vol. 14, No. 3, pp. 347-359, 2025. doi: 10.18178/ijmerr.14.3.347-359

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).