EGT: Execution-Grounded Traversability through Learned Closed-Loop Trajectory Distributions
EGT: Execution-Grounded Traversability through Learned Closed-Loop Trajectory Distributions
This paper introduces Execution-Grounded Traversability (EGT), a method that estimates the probability of successfully executing a path by predicting closed-loop trajectory distributions. The approach uses a latent-space flow matching model conditioned on terrain features and vehicle state to improve traversability estimation over traditional point-wise metrics. Evaluated across 650 simulated trials on unseen terrains, EGT reduces safe-path rejection to 2.2% compared to baselines ranging from 11.4% to 39.5%, and unsafe acceptance to 7.7%. When integrated into a receding-horizon sampling-based planner, it reduces navigation failure rates by 24-42 and 30-44 percentage points over prior methods, while cutting goal-reaching time by up to 20.5% and 45%. The authors also demonstrate zero-shot transfer of the simulation-trained model to real-world outdoor experiments.
The authors introduce Execution-Grounded Traversability (EGT), a method that estimates the probability of successfully executing a path by predicting closed-loop trajectory distributions, demonstrating zero-shot transfer from simulation to real-world outdoor experiments with reduced safe-path rejection and navigation failure rates.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
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 zer
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org