Social navigation adapts online with performance preserved by diffusion steering
Performance-Preserving Online Adaptation in Social Navigation via Diffusion Steering
Robotics
Summary
Navigating around people is tricky for robots because social behaviors and robot movements vary a lot and are hard to simulate perfectly. The authors propose a way for robots to learn and adapt to new social environments without losing their original ability to avoid people and get to their destination. Their method uses a special kind of learning called diffusion steering where only a small part of the robot’s behavior is updated, keeping the core skill intact. Tests show this allows robots to learn safely and flexibly, even on a real robot connected to simulations.
What this means in practice
- •For robot control engineers: Improve robot navigation systems by adapting behavior online to new social conventions without losing navigation safety.
- •For autonomous vehicle developers: Enable adaptive pedestrian avoidance in self-driving cars through learned updating of social navigation policies while preserving core safety.
Authors
Haruto Nagahisa, Kohei Matsumoto, Yuki Hyodo, Ryo Kurazume
Abstract
In social navigation, modeling the complex interactions between humans and robots is difficult, and deep reinforcement learning has therefore been actively studied. However, because simulation alone cannot fully reproduce diverse scenarios, robot dynamics, and the social conventions that vary across deployment environments, fine-tuning in the deployment environment is promising. In doing so, learning that preserves the base model's performance is required, so as not to compromise the primary objective of navigation, namely avoiding pedestrians and reaching the destination. In this study, we propose a method that applies diffusion steering via reinforcement learning (DSRL), which trains only the noise policy while keeping the diffusion policy fixed, thereby achieving learning that preserves performance. Furthermore, we integrate diffusion-based RL policies trained with multiple seeds to construct the base policy, improving learning performance. Our evaluation shows that, compared with other methods, the proposed method enables efficient learning while preserving performance, and we confirm flexible behavior control through adaptation to social conventions, as well as its effectiveness on a physical robot through hardware-in-the-loop simulation.