Trajectory-Diversity-Driven Robust Vision-and-Language Navigation

2026-03-16

AI Summary

This work addresses the poor generalization and low robustness to execution perturbations in imitation learning for vision-and-language navigation by proposing NavGRPO, a novel framework that introduces Group Relative Policy Optimization. NavGRPO drives policy updates through intra-group trajectory performance comparisons without requiring an auxiliary value network. By leveraging trajectory diversity to encourage exploration, the method reduces reliance on expert demonstrations and significantly enhances robustness in unseen environments and under perturbations. Experimental results demonstrate consistent improvements on the R2R and REVERIE benchmarks, with Success weighted by Path Length (SPL) gains of 3.0% and 1.71%, respectively. Notably, under severe early-stage perturbations, NavGRPO achieves an SPL improvement of up to 14.89%.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Intelligent Robots: Learning & Optimization for ROB
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Vision-and-Language Navigation (VLN) requires agents to navigate photo-realistic environments following natural language instructions. Current methods predominantly rely on imitation learning, which suffers from limited generalization and poor robustness to execution perturbations. We present NavGRPO, a reinforcement learning framework that learns goal-directed navigation policies through Group Relative Policy Optimization. By exploring diverse trajectories and optimizing via within-group performance comparisons, our method enables agents to distinguish effective strategies beyond expert paths without requiring additional value networks. Built on ScaleVLN, NavGRPO achieves superior robustness on R2R and REVERIE benchmarks with +3.0% and +1.71% SPL improvements in unseen environments. Under extreme early-stage perturbations, we demonstrate +14.89% SPL gain over the baseline, confirming that goal-directed RL training builds substantially more robust navigation policies. Code and models will be released.

Problem

• Vision-and-Language Navigation
• imitation learning
• robustness
• generalization
• execution perturbations

Innovation

• trajectory diversity
• group relative policy optimization
• robust vision-and-language navigation
• reinforcement learning
• goal-directed navigation