End-to-End Safe Social Navigation via Multi-Task Reinforcement Learning and Probabilistic Perception

2026-10-05

AI Summary

This study addresses the challenge of simultaneously achieving efficiency, safety, and social compliance in autonomous social navigation, as well as the reliance of existing reinforcement learning methods on human-state priors. To this end, we propose JESSI, an end-to-end framework that directly maps raw LiDAR observations to control commands via multi-task reinforcement learning. JESSI parameterizes a continuous action space using Dirichlet distributions with deterministic boundaries to guarantee physical safety, and incorporates attention mechanisms to extract probabilistic human states for interpretable, socially aware decision-making. Both simulation and real-world robot experiments demonstrate that JESSI jointly optimizes multiple performance metrics without requiring manual priors, significantly outperforming baseline methods in navigation success rate and social compliance.

Abstract

Autonomous social navigation requires balancing efficiency, physical safety, and social compliance. Reinforcement Learning (RL) methods provide a viable and effective solution but often rely on unrealistic assumptions, such as the knowledge of humans'position and velocity. In this paper, we introduce JESSI (JAX-based E2E Safe Social Interpretable navigation), a lightweight end-to-end RL framework that maps raw LiDAR scans directly to kinematically feasible control commands. JESSI enhances safety via Dirichlet-parameterized continuous action spaces and deterministic bounding, while an integrated attention-based perception module extracts probabilistic human states for interpretable, socially aware decision-making. Through extensive simulations and real-world deployment on a differential-drive robot, we demonstrate that jointly optimizing the RL policy with a supervised perception signal in a multi-task paradigm enhances social behavior. Ultimately, JESSI is able to balance high navigation success rates and superior social behaviors compared to state-of-the-art baselines.

Problem

• Social Navigation
• Reinforcement Learning
• End-to-End
• Safety
• Probabilistic Perception

Innovation

• End-to-End Reinforcement Learning
• Social Navigation
• Probabilistic Perception
• Multi-Task Learning
• Safe Action Space