COLSON: Controllable Learning-Based Social Navigation via Diffusion-Based Reinforcement Learning

In three linesCOLSON applies diffusion-based reinforcement learning to social navigation for mobile robots in dynamic environments. The method outperforms Gaussian policy approaches in action flexibility and adapts to unseen scenarios (static obstacles, variable objectives) without retraining.

Summary generated by Claude — human-verified