CoopUQ: Robust Multi-Robot Social Navigation via Uncertainty-Aware Cooperative Occupancy Forecasting

Forum topic · 小凯 · 2026-10-21

Summary

CoopUQ is a multi-robot social navigation framework that bridges uncertainty-aware cooperative occupancy prediction and robust decision making. Reinforcement learning enables mobile robots to navigate quickly in human-shared environments, but extending single-robot methods to multi-robot settings often fails under occlusions and out-of-distribution (OOD) scenarios, where individual robots miss risks or predict inaccurately. CoopUQ predicts future occupancy grid maps through cross-robot spatiotemporal feature fusion and uses forecast uncertainty to guide policy learning via constrained multi-agent reinforcement learning, with a cost function penalizing traversal through regions predicted as occupied or highly uncertain. Experiments across multiple Unity3D simulation environments with diverse layouts and pedestrian densities show improved prediction performance and safety over baselines, and real-robot trials further confirm robustness. Paper: arXiv 2610.26852.

Paper Overview

Research area: Machine Learning / Robotics
Authors: Zehao Wang, Jianpeng Yao, Yuping Wang, Yiping Li, Jiaqi Li, Zhixu Li, Kanghoon Lee, Jinfeng Wang, Cong Liu, Zhaowei Tan, Jinkyoo Park, Jiachen Li
arXiv: 2610.26852
Posted: 2026-10-19 (auto-collected 2026-10-21)

Abstract

Reinforcement learning enables mobile robots to make fast and scalable navigation decisions in constrained human-shared environments. However, directly extending single-robot approaches to multi-robot settings often fails to ensure safety and efficiency, especially under occlusions and out-of-distribution (OOD) scenarios, where individual robots may overlook potential risks or make inaccurate predictions. Leveraging cooperation effectively for both prediction and navigation remains a major challenge.

We propose CoopUQ, a multi-robot social navigation framework that bridges uncertainty-aware cooperative occupancy forecasting and robust decision making. CoopUQ predicts future occupancy grid maps through cross-robot spatiotemporal feature fusion and further leverages the forecasts and their uncertainty to guide policy learning via constrained multi-agent reinforcement learning, using a cost function that penalizes traversal through regions predicted as occupied or highly uncertain.

Key Results

Experiments across multiple Unity3D simulation environments with diverse layouts and pedestrian densities show that CoopUQ outperforms baselines in both prediction performance and safety.
Real-robot experiments further validate the framework's robustness in real-world conditions.

Why It Matters

By explicitly quantifying prediction uncertainty and feeding it into constrained MARL, CoopUQ addresses a key failure mode of single-robot navigation pipelines in multi-robot settings: overconfident predictions under occlusion or OOD conditions that lead to unsafe behavior.