Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Probabilistic Traversability with Test-time Adaptation for Uncertainty-aware Planetary Rover Navigation

Masafumi Endo, Tatsunori Taniai, Genya Ishigami

TL;DR

This work integrates principal approaches to uncertainty handling -- quantification, exploitation, and adaptation -- into a single learning and planning framework for rover navigation, forming the basis of an end-to-end probabilistic ML model that predicts slip distributions directly from rover traverse observations.

Abstract

Traversability assessment of deformable terrain is vital for safe rover navigation on planetary surfaces. Machine learning (ML) is a powerful tool for traversability prediction but faces predictive uncertainty. This uncertainty leads to prediction errors, increasing the risk of wheel slips and immobilization for planetary rovers. To address this issue, we integrate principal approaches to uncertainty handling -- quantification, exploitation, and adaptation -- into a single learning and planning framework for rover navigation. The key concept is \emph{deep probabilistic traversability}, forming the basis of an end-to-end probabilistic ML model that predicts slip distributions directly from rover traverse observations. This probabilistic model quantifies uncertainties in slip prediction and exploits them as traversability costs in path planning. Its end-to-end nature also allows adaptation of pre-trained models with in-situ traverse experience to reduce uncertainties. We perform extensive simulations in synthetic environments that pose representative uncertainties in planetary analog terrains. Experimental results show that our method achieves more robust path planning under novel environmental conditions than existing approaches.

Deep Probabilistic Traversability with Test-time Adaptation for Uncertainty-aware Planetary Rover Navigation

TL;DR

This work integrates principal approaches to uncertainty handling -- quantification, exploitation, and adaptation -- into a single learning and planning framework for rover navigation, forming the basis of an end-to-end probabilistic ML model that predicts slip distributions directly from rover traverse observations.

Abstract

Traversability assessment of deformable terrain is vital for safe rover navigation on planetary surfaces. Machine learning (ML) is a powerful tool for traversability prediction but faces predictive uncertainty. This uncertainty leads to prediction errors, increasing the risk of wheel slips and immobilization for planetary rovers. To address this issue, we integrate principal approaches to uncertainty handling -- quantification, exploitation, and adaptation -- into a single learning and planning framework for rover navigation. The key concept is \emph{deep probabilistic traversability}, forming the basis of an end-to-end probabilistic ML model that predicts slip distributions directly from rover traverse observations. This probabilistic model quantifies uncertainties in slip prediction and exploits them as traversability costs in path planning. Its end-to-end nature also allows adaptation of pre-trained models with in-situ traverse experience to reduce uncertainties. We perform extensive simulations in synthetic environments that pose representative uncertainties in planetary analog terrains. Experimental results show that our method achieves more robust path planning under novel environmental conditions than existing approaches.
Paper Structure (23 sections, 11 equations, 4 figures, 2 tables)

This paper contains 23 sections, 11 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Path Planning
  5. Traversability as Wheel Slip
  6. Map Representation
  7. Uncertainty-aware Navigation Algorithm
  8. Deep Probabilistic Traversability Model
  9. Uncertainty-aware Cost Formulation
  10. Test-time Domain Adaptation
  11. Simulation Experiments
  12. Dataset
  13. Overall data generation
  14. In-domain data generation
  15. Out-of-domain test data generation
  16. ...and 8 more sections

Figures (4)

  • Figure 1: Overview of the proposed unified framework for planetary rover navigation, consisting of (a) deep probabilistic traversability, (b) uncertainty-aware path planning, and (c) test-time DA with in-situ slip measurements, forming an iterative loop to address uncertain traversability prediction.
  • Figure 2: Dataset visualizations. (a) Ten latent slip functions associated with terrain classes, with shaded areas indicating the $2\sigma$ ranges of their additive noises. (b) Distribution of terrain slope angles in the training subset. The gray shading in (a) and (b) indicates the sampling range of the crater slopes in (d) and (f). (c)-(f) Example color maps for the in-domain, unfamiliar geometry, unfamiliar appearance, and unfamiliar geometry and appearance test subsets.
  • Figure 3: Traversability predictions and navigation results for a map in the unfamiliar geometry and appearance subset. The left column shows the ground-truth slips (top) and slope angles (bottom) for the rightward movements. The middle columns display slip predictions before (top row) and after (bottom row) domain adaptation, showing absolute errors with total, aleatoric, and epistemic uncertainties. The right column shows path planning and execution results. The cyan square and yellow star indicate the start and goal locations, respectively; white crosses mark locations where rovers failed to traverse.
  • Figure 4: Parameter study for the unfamiliar geometry and appearance subset, showing the variations of five metrics with varying $\lambda$. The left percentile y-axis represents Sol, Suc, Sol-Suc, and $s_{\text{max}}$, while the right y-axis represents $T_{\text{total}}$. Blue and orange violin plots represent the distribution of $s_{\text{max}}$ and $T_{\text{total}}$, respectively, with lines marking the 25th, 50th, and 75th percentiles.