EGT: Execution-Grounded Traversability through Learned Closed-Loop Trajectory Distributions

Chinmay Mundane; Amith Manoharan; Arun Kumar Singh

cs.RONov 6, 2026

EGT: Execution-Grounded Traversability through Learned Closed-Loop Trajectory Distributions

Authors: Chinmay Mundane, Amith Manoharan, Arun Kumar Singh

Organizations: are with the Institute of Technology, University of Tartu, Estonia.

Abstract

Rough terrain traversability is often estimated through point-wise roll, pitch, or force-angle margins along a planned path. However, these metrics are often inaccurate and overly conservative because they ignore how terrain disturbances and low-level controller corrections can affect the vehicle's motions. To overcome these limitations, we introduce Execution-Grounded Traversability (EGT), defined as the probability that a vehicle-controller pair can execute a candidate path while meeting trajectory-level tracking and progress requirements. We estimate EGT by predicting the distribution of closed-loop trajectories using a latent-space flow matching (FM) model conditioned on multi-resolution terrain features and 3D vehicle attitude and wheel-ground contact points prior. Across 650 simulated trials on unseen terrains, EGT reduces safe-path rejection to 2.2% (versus 11.4%-39.5% for pose-margin and force-angle baselines) and unsafe acceptance to 7.7%. Integrated into a receding-horizon sampling-based planner, EGT reduces navigation failure rates by 24-42 and 30-44 percentage points over [1] and [2], respectively, while cutting goal-reaching time by up to 20.5% and 45%. Finally, we demonstrate zero-shot transfer of our simulation-trained model to real-world outdoor experiments.

Figures & tables

Fig. 2 : Overview of the conditional latent flow matching architecture. Terrain FPN and path-fusion features, kinematic pose–contact features, and planner–goal context form conditioning tokens $\mathbf{C}$ . A shared flow transformer predicts the conditional latent velocity field $v_{\theta}$ , which is integrated and decoded into closed-loop trajectory rollouts.

Panel of Figure 2 — Fig. 3 : Representative closed-loop trajectory predictions on simulated terrain.

Metric	Force-Angle [ 3 ]	Pose-Margin			EGT
		$45^{\circ}$	$60^{\circ}$	$75^{\circ}$	(Ours)	w/o pose- contact prior
Conservative rejection $\downarrow$	39.5	35.3	21.7	11.4	2.2	4.8
Unsafe acceptance $\downarrow$	13.5	13.5	38.5	63.5	7.7	71.2
Overall accuracy $\uparrow$	62.6	66.5	76.9	84.5	97.4	89.8

TABLE I : Traversability classification over 650 simulated trials: 598 safe and 52 unsafe executions under the stated task requirements. All metric values are percentages. The EGT ablation omits conditioning on the NLS pose prior. Arrows indicate the preferred direction, and best values are shown in bold. The force-angle baseline uses gravitational loading only.

Probability bin	$n_{b}$	$\bar{p}_{b}$	$\bar{y}_{b}$	$95\%$ CI
$[0,0.80)$	61	0.0656	0.2131	$[0.1290,0.3312]$
$[0.80,0.96)$	40	0.9200	1.0000	$[0.9124,1.0000]$
$[0.96,1)$	64	0.9667	1.0000	$[0.9434,1.0000]$
$p_{i}=1$	485	1.0000	0.9918	$[0.9790,0.9968]$

TABLE II : EGT probability calibration against the execution-success labels used in Table I .

n_{b}

: bin size;

\bar{p}_{b}

: mean predicted probability;

\bar{y}_{b}

: observed execution-success frequency; CI: pointwise

95\%

Wilson confidence interval for the group execution-success probability.

Terrain	Method	Failure rate (%) $\downarrow$	Goal time (s) $\downarrow$ Mean $\pm$ std.
Terrain 1	Proposed	4	$\mathbf{27.42}\pm 13.09$
	GP-navigation	40	$28.00\pm 14.46$
	TAL	34	$49.80\pm 22.22$
Terrain 2	Proposed	4	$\mathbf{32.90}\pm 10.65$
	GP-navigation	46	$37.68\pm 20.69$
	TAL	48	$51.37\pm 17.19$

TABLE III : Navigation performance across three terrains, with 50 trials per method and terrain. Failure rates use all trials; goal-reaching times use common-success runs (

N=21,17,19

for Terrains 1–3).

Robot Traversability Estimation

See all 31 papers

ProTrav: Probabilistic Traversability with Elevation Uncertainty Propagation for Risk-aware Off-road Navigation
Oct 22, 2026Chuanlin Zuo, Miloš Prágr, Arun Kumar Singh +2Robot Traversability Estimation Off-Road Robot Navigation
Faculty of Engineering and Natural Sciences, Tampere University, Finland. · Institute of Technology, University of Tartu, Estonia. · AI and ML Department, Volvo Autonomous Solutions, Sweden.
Learning Traversability for Long Horizon Off-Road Navigation
Jul 26, 2026Kasi Viswanath, Jason M. Gregory, Shaunak Kolhe +1Off-Road Robot Navigation Geospatial Representation Learning
J. Mike Walker ’66 Department of Mechanical Engineering, Texas A&M University, College Station, TX 77840, USA · DEVCOM Army Research laboratory, Adelphi, MD, USA
VCTP: Vehicle-Conditioned Terrain Planning for Off-Road Navigation
Sep 15, 2026Akshay Naik, Ramavarapu S. Sreenivas, Dustin Nottage +1Constrained Motion Planning Off-Road Robot Navigation
University of Illinois Urbana-Champaign, Urbana, IL, USA: 1Department of Electrical and Computer Engineering · University of Illinois Urbana-Champaign, Urbana, IL, USA: 2Department of Industrial and Enterprise Systems Engineering · U.S. Army Engineer Research and Development Center, Construction Engineering Research Laboratory, Champaign, IL, USA

Oct 22, 2026J/K move · Enter open · S save

ProTrav: Probabilistic Traversability with Elevation Uncertainty Propagation for Risk-aware Off-road Navigation

Chuanlin Zuo, Miloš Prágr, Arun Kumar Singh, Mohammad Loni+1

Faculty of Engineering and Natural Sciences, Tampere University, Finland. · Institute of Technology, University of Tartu, Estonia. · AI and ML Department, Volvo Autonomous Solutions, Sweden.

Robot Traversability Estimation Off-Road Robot Navigation Risk-Aware Path Planning

Open paper AskPDF arXiv

Reliable traversability estimation is essential for autonomous navigation on unstructured terrain. State-of-the-art approaches typically evaluate traversability from deterministic terrain elevation, thereby overlooking spatial correlations in the terrain and failing to account for how elevation uncertainty propagates through nonlinear traversability mappings. We propose ProTrav, a probabilistic terrain representation that uses function-space sampling to generate spatially correlated terrain realizations and propagate elevation uncertainty to traversability cost. First, a Sparse Variational Gaussian Process (SVGP) reconstructs a probabilistic elevation field from observations. Then, posterior function-space sampling generates spatially correlated terrain realizations, preserving spatial dependencies between neighboring locations while propagating elevation uncertainty through geometric features into traversability cost. Consequently, a distribution of traversability cost at arbitrary location is established, where the mean and standard deviation quantify the expected traversability cost and associated uncertainty, respectively. A risk term is defined by moment-based Upper Confidence Bound (UCB) and incorporated into a Model Predictive Path Integral (MPPI) controller for risk-aware navigation. Simulation experiments demonstrate navigation success rates exceeding 90% on different terrains, compared with below 40% for a baseline, while reducing attitude variations. Furthermore, the proposed approach is validated on a robot traversing real-world off-road terrain.

Abstract

Robot Traversability Estimation

Explore similar work

ProTrav: Probabilistic Traversability with Elevation Uncertainty Propagation for Risk-aware Off-road Navigation

Learning Traversability for Long Horizon Off-Road Navigation

VCTP: Vehicle-Conditioned Terrain Planning for Off-Road Navigation