Traverse the Non-Traversable: Estimating Traversability for Wheeled Mobility on Vertically Challenging Terrain

TL;DR

We address the challenge of navigating vertically challenging terrain with wheeled robots by moving beyond binary traversable/non-traversable maps. The Traverse the Non-Traversable (tnt) estimator uses past kinodynamic interactions to produce patch-level traversability, which is efficiently turned into a real-time traversability map that can guide both sampling-based planners with a high-precision 6-DoF kinodynamic model and costmap-based planners. TNT combines roll/pitch amplitude, velocity uncertainty, and pose-prediction uncertainty with learnable weights to estimate patch traversability, and reconstructs a map suitable for onboard planning. Experimental results on a 1/10-scale vehicle show substantial improvements in success rate, traversal time, and stability, with demonstrations in indoor, outdoor, and real-world rock terrains highlighting practical impact for autonomous off-road navigation.

Abstract

Most traversability estimation techniques divide off-road terrain into traversable (e.g., pavement, gravel, and grass) and non-traversable (e.g., boulders, vegetation, and ditches) regions and then inform subsequent planners to produce trajectories on the traversable part. However, recent research demonstrated that wheeled robots can traverse vertically challenging terrain (e.g., extremely rugged boulders comparable in size to the vehicles themselves), which unfortunately would be deemed as non-traversable by existing techniques. Motivated by such limitations, this work aims at identifying the traversable from the seemingly non-traversable, vertically challenging terrain based on past kinodynamic vehicle-terrain interactions in a data-driven manner. Our new Traverse the Non-Traversable(TNT) traversability estimator can efficiently guide a down-stream sampling-based planner containing a high-precision 6-DoF kinodynamic model, which becomes deployable onboard a small-scale vehicle. Additionally, the estimated traversability can also be used as a costmap to plan global and local paths without sampling. Our experiment results show that TNT can improve planning performance, efficiency, and stability by 50%, 26.7%, and 9.2% respectively on a physical robot platform.

Figures (5)

• Figure 1: Guided by our tnt traversability estimator, a sampling-based motion planner with a high-precision kinodynamic model quickly converges to safe trajectories on vertically challenging terrain (bottom, colored traversability map), whereas without the tnt traversability map the planner gets stuck at exploring completely non-traversable area (top, white boulder on the black-white elevation map).
• Figure 2: tnt Overview. Based on terrain patches $p$ on the elevation map $E_i$, three predictors produce roll and pitch angles , velocity uncertainty $(\mu^{\Delta \mathbf{v}}, \sigma^{\Delta \mathbf{v}})$, and pose prediction uncertainty $(\mu^{\Delta \mathbf{q}}, \sigma^{\Delta \mathbf{q}})$, which are combined by $\mathbf{w}_1$, $\mathbf{w}_2$, and $\mathbf{w}_3$ to generate patch-wise traversability values; All traversability values form a traversability map $\text{Tm}_i$, which a map-wise traversability map estimator ($e_\eta(\cdot)$) learns to reconstruct based on terrain patch $E_i$.
• Figure 3: Roll and Pitch Model.
• Figure 4: Traversability Map Generated by tnt. The color gradient represents traversability, with blue areas indicating easily traversable terrain and red areas signifying challenging or non-traversable regions. The overlaid path represents the optimal route calculated by the A* algorithm.
• Figure 5: tnt Indoor Experiments and Outdoor Demonstration.