Online Adaptive Traversability Estimation through Interaction for Unstructured, Densely Vegetated Environments

Fabio A.Ruetz1,2, Nicholas Lawrance2, Emili Hernández3, Paulo V. K. Borges4, Thierry Peynot1
1 Queensland Centre for Robotics, University of Technology (QUT)
2 Commonwealth Scientific and Industrial Research Organisation (CSIRO)
3 Emesent
 4 Orica
Paper Code arXiv

Abstract

Navigating densely vegetated environments poses significant challenges for autonomous ground vehicles. Learning-based systems typically use prior and in-situ data to predict terrain traversability but often degrade in performance when encountering out-of-distribution elements caused by rapid environmental changes or novel conditions. This paper presents a novel, lidar-only, online adaptive traversability estimation (TE) method that trains a model directly on the robot using self-supervised data collected through robot-environment interaction. The proposed approach utilises a probabilistic 3D voxel representation to integrate lidar measurements and robot experience, creating a salient environmental model. To ensure computational efficiency, a sparse graph-based representation is employed to update temporarily evolving voxel distributions. Extensive experiments with an unmanned ground vehicle in natural terrain demonstrate that the system adapts to complex environments with as little as 8 minutes of operational data, achieving a Matthews Correlation Coefficient (MCC) score of 0.63 and enabling safe navigation in densely vegetated environments. This work examines different training strategies for voxel-based TE methods and offers recommendations for training strategies to improve adaptability. The proposed method is validated on a robotic platform with limited computational resources (25W GPU), achieving accuracy comparable to offline-trained models while maintaining reliable performance across varied environments.

Method Overview

The overview of the online, adaptive Traversability Estiamton method. The ForestTrav method (blue and yellow blocks) is augmented with new modules to facilitate online learning (green). The probabilistic 3D voxel map, generated from lidar measurements, is fused with the robots experience, the probabilistic collision map. A sparse graph is used to maintain and update local map patches in a efficient and computational trackable manners. These patches are used to train and update models online, on the robot. The model weights of the TE node are update.

Online Adaptive Traversability Estimation Concept in Action

Navigation of Model Trained Purely Online

A new model (random weights) was trained online, on the platform in target environment. The model was trained on a Jetson Orin Nano (25W) within 8 minutes, using only data collected in situ. The following navigation is uncut footage of the navigation.

Additional Examples of Navigation of the ForestTrav method

BibTeX


        article{ruetz2025online,
        title={Online Adaptive Traversability Estimation through Interaction for Unstructured, Densely Vegetated Environments},
        author={Ruetz, Fabio A and Lawrance, Nicholas and Hern{\'a}ndez, Emili and Borges, Paulo VK and Peynot, Thierry},
        journal={arXiv preprint arXiv:2502.01987},
        year={2025}
      }