SaferPath: Hierarchical Visual Navigation with Learned Guidance and Safety-Constrained Control
Lingjie Zhang, Zeyu Jiang, Changhao Chen
AI summary
Problem
End-to-end visual navigation models struggle to generalize and ensure safety in unseen, cluttered, or narrow indoor environments, often leading to collisions due to distribution mismatches and lack of dense training data.
Approach
SaferPath uses a hierarchical framework that takes coarse trajectories from an end-to-end model and refines them in real-time using a novel Model Predictive Stein Variational Evolution Strategy optimized against a traversability score map, followed by MPC tracking.
Key results
- Introduces SaferPath, a modular hierarchical framework integrating learned guidance with safety-constrained control.
- Proposes MP-SVES, an efficient optimization algorithm generating safe trajectories in ~10 iterations.
- Achieves 40% higher success rate under unseen obstacles and >50% improvement in dense unstructured exploration.
- Successfully navigates narrow corridors and real-world settings where baselines fail.
Why it matters
Enables reliable, collision-free autonomous navigation for indoor robots in complex, dynamic environments without requiring model retraining.
Abstract
Visual navigation is a core capability for mobile robots, yet end-to-end learning-based methods often struggle with generalization and safety in unseen, cluttered, or narrow environments. These limitations are especially pronounced in dense indoor settings, where collisions are likely and end-to-end models frequently fail. To address this, we propose SaferPath, a hierarchical visual navigation framework that leverages learned guidance from existing end-to-end models and refines it through a safety-constrained optimization-control module. SaferPath transforms visual observations into a traversable-area map and refines guidance trajectories using Model Predictive Stein Vari- ational Evolution Strategy (MP-SVES), efficiently generating safe trajectories in only a few iterations. The refined trajectories are tracked by an MPC controller, ensuring robust navigation in complex environments. Extensive experiments in scenarios with unseen obstacles, dense unstructured spaces, and narrow corridors demonstrate that SaferPath consistently improves success rates and reduces collisions, outperforming represen- tative baselines such as ViNT and NoMaD, and enabling safe navigation in challenging real-world settings.