MetricNet: Recovering Metric Scale in Generative Navigation Policies
Abhijeet Kishore Nayak, Débora Oliveira Makowski, Samiran Gode, Cordelia Schmid, Wolfram Burgard
AI summary
Problem
Generative navigation policies output abstract, unscaled waypoints that lack metric grounding, leading to short-sighted, unsafe actions and requiring manual scaling for deployment.
Approach
MetricNet predicts the real-world metric scale factor for waypoints sampled by diffusion-based navigation policies, grounding them in executable metric coordinates, while MetricNav uses these scaled points for simultaneous goal-following and collision avoidance.
Key results
- Accurate prediction of metric scale factors for generative policy outputs
- Significant improvement in navigation and exploration performance in simulation
- MetricNav successfully guides trajectories away from obstacles while maintaining goal-directed movement
- Successful real-world robot deployment without manual tuning
Why it matters
Enables robust, safe, and generalizable deployment of end-to-end generative navigation policies on real robots without environment-specific tuning.
Abstract
Generative navigation policies have made rapid progress in improving end-to-end learned navigation. Despite their promising results, this paradigm has two structural problems. First, the sampled trajectories exist in an abstract, unscaled space without metric grounding. Second, the control strategy discards the full path, instead moving directly towards a single waypoint. This leads to short-sighted and unsafe actions, moving the robot towards obstacles that a complete and correctly scaled path would circumvent. To address these issues, we propose MetricNet, an effective add-on for generative navigation that predicts the metric distance between waypoints, grounding policy outputs in metric coordinates. We evaluate our method in simulation with a new benchmarking framework and show that executing MetricNet-scaled waypoints significantly improves both navigation and exploration performance. Beyond simulation, we further validate our approach in real-world experiments. Finally, we propose MetricNav, which integrates MetricNet into a navigation policy to guide the robot away from obstacles while still moving towards the goal.