Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dynamics Models in the Aggressive Off-Road Driving Regime

Tyler Han, Sidharth Talia, Rohan Panicker, Preet Shah, Neel Jawale, Byron Boots

TL;DR

The paper tackles the challenge of modeling dynamics for aggressive off-road driving, where large inertial effects require accurate higher-order state predictions for safe MPC. It compares three dynamics model classes—SE(3) no-slip bicycle, SE(3) slip-based, and fully learned terrain-conditioned dynamics—on BeamNG simulation data and real-world Washington data, using the horizon-aggregated $H$-MNE and a free-energy measure $E(\tau)$ to quantify aggressiveness. Results show that model accuracy degrades with increasing aggressiveness, with simpler models deteriorating more rapidly, and cross-dataset validation highlights the need for safety-critical evaluation. The work provides benchmarks and introduces an aggressiveness metric that can guide future developments, including incorporating terrain semantics via terrainnet to improve robustness in real-world aggressive off-road scenarios.

Abstract

Current developments in autonomous off-road driving are steadily increasing performance through higher speeds and more challenging, unstructured environments. However, this operating regime subjects the vehicle to larger inertial effects, where consideration of higher-order states is necessary to avoid failures such as rollovers or excessive impact forces. Aggressive driving through Model Predictive Control (MPC) in these conditions requires dynamics models that accurately predict safety-critical information. This work aims to empirically quantify this aggressive operating regime and its effects on the performance of current models. We evaluate three dynamics models of varying complexity on two distinct off-road driving datasets: one simulated and the other real-world. By conditioning trajectory data on higher-order states, we show that model accuracy degrades with aggressiveness and simpler models degrade faster. These models are also validated across datasets, where accuracies over safety-critical states are reported and provide benchmarks for future work.

Dynamics Models in the Aggressive Off-Road Driving Regime

TL;DR

The paper tackles the challenge of modeling dynamics for aggressive off-road driving, where large inertial effects require accurate higher-order state predictions for safe MPC. It compares three dynamics model classes—SE(3) no-slip bicycle, SE(3) slip-based, and fully learned terrain-conditioned dynamics—on BeamNG simulation data and real-world Washington data, using the horizon-aggregated -MNE and a free-energy measure to quantify aggressiveness. Results show that model accuracy degrades with increasing aggressiveness, with simpler models deteriorating more rapidly, and cross-dataset validation highlights the need for safety-critical evaluation. The work provides benchmarks and introduces an aggressiveness metric that can guide future developments, including incorporating terrain semantics via terrainnet to improve robustness in real-world aggressive off-road scenarios.

Abstract

Current developments in autonomous off-road driving are steadily increasing performance through higher speeds and more challenging, unstructured environments. However, this operating regime subjects the vehicle to larger inertial effects, where consideration of higher-order states is necessary to avoid failures such as rollovers or excessive impact forces. Aggressive driving through Model Predictive Control (MPC) in these conditions requires dynamics models that accurately predict safety-critical information. This work aims to empirically quantify this aggressive operating regime and its effects on the performance of current models. We evaluate three dynamics models of varying complexity on two distinct off-road driving datasets: one simulated and the other real-world. By conditioning trajectory data on higher-order states, we show that model accuracy degrades with aggressiveness and simpler models degrade faster. These models are also validated across datasets, where accuracies over safety-critical states are reported and provide benchmarks for future work.
Paper Structure (13 sections, 6 equations, 4 figures, 1 table)

This paper contains 13 sections, 6 equations, 4 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. Literature Review
  3. Dynamics models
  4. SE(3) no-slip based bicycle model
  5. SE(3) slip based approximate model
  6. Terrain conditioned dynamics modeling
  7. Aggressiveness
  8. Definition
  9. Measuring Aggressiveness with Free Energy
  10. Experimentation
  11. Aggressiveness and Dynamics Models
  12. Model Learning with Aggressive Data
  13. Conclusion

Figures (4)

  • Figure 1: This work evaluates dynamics models of varying complexity in challenging and unstructured environments, both simulated (BeamNG, left) and real (Washington, right).
  • Figure 2: Illustration of the coordinate frame used by dynamics models
  • Figure 3: Freeze-frames of aggressive trajectories in BeamNGRL dataset.
  • Figure 4: Model performance on validation trajectories. Each marker corresponds to the validation of a model on a single trajectory. A cubic polynomial is fit to each model's performance to show the general trend over increasing Free Energy Values $E(\tau)$. Higher energy indicates more out-of-distribution (more aggressive) relative to the initial dataset.