Pushing the Limits of Reactive Navigation: Learning to Escape Local Minima
Isar Meijer, Michael Pantic, Helen Oleynikova, Roland Siegwart
AI summary
Problem
Purely reactive navigation methods are computationally efficient and map-free but frequently get trapped in local minima, while map-based planners avoid this but require explicit environmental representations. This work asks whether reactive navigation can be augmented with learned geometric intuition to escape dead-ends without relying on a map.
Approach
The authors combine a classical reactive obstacle-avoidance algorithm with feed-forward and recurrent neural networks trained in self-supervised environments. The networks learn to predict optimal goal directions using only local sensor rays, effectively giving the robot geometric intuition to bypass dead-ends.
Key results
- Zero-shot transfer to real-world and synthetic 3D environments
- Robust navigation under up to 30% sensor noise
- Recurrent networks outperform feed-forward models in escaping local minima
- Open-source release of planners and experimental code
Why it matters
This work bridges the gap between fast reactive navigation and global planning, offering a practical, map-free solution for autonomous robots operating in complex, cluttered environments.
Abstract
Can a robot navigate a cluttered environment with- out an explicit map? Reactive methods that use only the robot’s current sensor data and local information are fast and flexible, but prone to getting stuck in local minima. Is there a middle- ground between reactive methods and map-based path plan- ners? In this paper, we investigate feed forward and recurrent networks to augment a purely reactive sensor-based navigation algorithm, which should give the robot “geometric intuition” about how to escape local minima. We train on a large number of extremely cluttered simulated worlds, auto-generated from primitive shapes, and show that our system zero-shot transfers to worlds based on real data, 3D man-made environments, and can handle up to 30% sensor noise without degradation of performance. We also offer a discussion of what role network memory plays in our final system, and what insights can be drawn about the nature of reactive vs. map-based navigation. The implementation of the planners and all experiments is made available open-source https://github.com/ethz-asl/rmp dl.