Self-Supervised Online Robot-Agnostic Traversability Estimation for Open-World Environments

2026-05-27

AI Summary

Existing approaches struggle to achieve robot-agnostic, online, and low-overhead traversability estimation in open-world settings, hindering safe and efficient navigation in unknown environments. This work proposes COTRATE, a framework that leverages self-supervised online learning to fuse proprioceptive and inertial signals into a robot-agnostic traversability score, which in turn guides a visual network to continuously adapt to novel terrains. Key innovations include a robot-agnostic online terrain assessment module, a vision–terrain alignment loss function, and a diversity-aware compact replay memory strategy. Evaluated on a large-scale dataset encompassing 11 outdoor terrain types, 50,000 images, and two robotic platforms, COTRATE significantly enhances navigation performance, enables cross-platform transferability, and substantially reduces computational and storage overhead. The code, models, and dataset are publicly released.

Abstract

Self-supervised online traversability estimation enables robots to continuously learn from unlabeled open-world experiences and adapt their navigation behavior toward safe and efficient trajectories. Existing approaches either rely on handcrafted proprioceptive traversability scores, limiting robot-agnosticism, or cluster prior data, preventing online learning. Moreover, many continual learning methods incur substantial memory and computational costs, hindering onboard deployment. We introduce COTRATE, an online learning framework for continuous traversability estimation from multimodal, unlabeled robot experience. Our method first infers robust traversability scores using a robot-agnostic, learning-based online terrain assessment module operating on proprioceptiveand inertial signals. These scores then supervise a visual traversability network through a novel alignment loss that associates visual embeddings with online terrain assessments.To mitigate forgetting during continual learning with minimal overhead, we propose a diversity-aware feature selection strategythat preserves performance using a compact replay memory. We further show that the learned traversability representation supports knowledge transfer across different robot platforms with different locomotion kinematics. We evaluate COTRATE on a dataset of \approx 50,000 images collected with two robotic platforms across 11 outdoor terrains, and benchmark it on navigation tasks in three representative outdoor environments. We make the dataset, code, and trained models publicly available.

Problem

• traversability estimation
• self-supervised learning
• robot-agnostic
• online learning
• open-world environments

Innovation

• self-supervised learning
• robot-agnostic traversability
• online continual learning
• multimodal perception
• knowledge transfer