Learning to Drive Anywhere with Model-Based Reannotation
Noriaki Hirose, Lydia Ignatova, Kyle Stachowicz, Catherine Glossop, Sergey Levine, Dhruv Shah
AI summary
Problem
Training broadly generalizable robot navigation policies is bottlenecked by the scarcity of large-scale, high-quality datasets, while abundant passive data sources are too noisy or lack action labels for direct imitation learning.
Approach
The authors introduce Model-Based ReAnnotation (MBRA), which uses a short-horizon expert model to generate clean action labels for noisy or action-free passive data, then distills these labels into a long-horizon navigation policy called LogoNav.
Key results
- Successful relabeling of noisy crowd-sourced and action-free video data
- State-of-the-art zero-shot navigation over 300+ meters in unseen environments
- Robust real-world deployment across a global robot fleet in six cities
- Superior performance over inverse-model and multi-step goal-conditioned baselines
Why it matters
Demonstrates that cheap, abundant passive data can replace costly expert datasets for training globally generalizable robot navigation policies.
Abstract
Developing broadly generalizable visual naviga- tion policies for robots is a significant challenge, primarily constrained by the availability of large-scale, diverse training data. While curated datasets collected by researchers offer high quality, their limited size restricts policy generalization. To overcome this, we explore leveraging abundant, passively collected data sources, including large volumes of crowd- sourced teleoperation data and unlabeled YouTube videos, despite their potential for lower quality or missing action labels. We propose Model-Based ReAnnotation (MBRA), a framework that utilizes a learned short-horizon, model-based expert model to relabel or generate high-quality actions for these passive datasets. This relabeled data is then distilled into LogoNav, a long-horizon navigation policy conditioned on visual goals or GPS waypoints. We demonstrate that LogoNav, trained using MBRA-processed data, achieves state-of-the-art performance, enabling robust navigation over distances exceeding 300 meters in previously unseen indoor and outdoor environments. Our extensive real-world evaluations, conducted across a fleet of robots (including quadrupeds) in six cities on three continents, validate the policy’s ability to generalize and navigate effectively even amidst pedestrians in crowded settings.