Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Model Validity in Observers: When to Increase the Complexity of Your Model?

Agapius Bou Ghosn, Philip Polack, Arnaud de La Fortelle

TL;DR

This work analyzes when vehicle dynamics representations remain valid by comparing four common model configurations against real-vehicle data and identifying a lateral-acceleration boundary near $a_y=0.5g$ that separates accurate from degraded performance. It shows Pacejka-based tire formulations yield the most robust state evolution and observer performance in high-dynamics, while linear and Dugoff tires exhibit larger errors; a learned observer can outperform all model-based approaches in high-$a_y$ regimes. The study highlights the practical need to match the modeling choice to the expected maneuver envelope in planning and control to ensure safety and accuracy. Overall, the results advocate domain-aware model selection and demonstrate the potential of learned observers to extend validity beyond traditional model limits.

Abstract

Model validity is key to the accurate and safe behavior of autonomous vehicles. Using invalid vehicle models in the different plan and control vehicle frameworks puts the stability of the vehicle, and thus its safety at stake. In this work, we analyze the validity of several popular vehicle models used in the literature with respect to a real vehicle and we prove that serious accuracy issues are encountered beyond a specific lateral acceleration point. We set a clear lateral acceleration domain in which the used models are an accurate representation of the behavior of the vehicle. We then target the necessity of using learned methods to model the vehicle's behavior. The effects of model validity on state observers are investigated. The performance of model-based observers is compared to learning-based ones. Overall, the presented work emphasizes the validity of vehicle models and presents clear operational domains in which models could be used safely.

Model Validity in Observers: When to Increase the Complexity of Your Model?

TL;DR

This work analyzes when vehicle dynamics representations remain valid by comparing four common model configurations against real-vehicle data and identifying a lateral-acceleration boundary near that separates accurate from degraded performance. It shows Pacejka-based tire formulations yield the most robust state evolution and observer performance in high-dynamics, while linear and Dugoff tires exhibit larger errors; a learned observer can outperform all model-based approaches in high- regimes. The study highlights the practical need to match the modeling choice to the expected maneuver envelope in planning and control to ensure safety and accuracy. Overall, the results advocate domain-aware model selection and demonstrate the potential of learned observers to extend validity beyond traditional model limits.

Abstract

Model validity is key to the accurate and safe behavior of autonomous vehicles. Using invalid vehicle models in the different plan and control vehicle frameworks puts the stability of the vehicle, and thus its safety at stake. In this work, we analyze the validity of several popular vehicle models used in the literature with respect to a real vehicle and we prove that serious accuracy issues are encountered beyond a specific lateral acceleration point. We set a clear lateral acceleration domain in which the used models are an accurate representation of the behavior of the vehicle. We then target the necessity of using learned methods to model the vehicle's behavior. The effects of model validity on state observers are investigated. The performance of model-based observers is compared to learning-based ones. Overall, the presented work emphasizes the validity of vehicle models and presents clear operational domains in which models could be used safely.

Paper Structure

This paper contains 23 sections, 19 equations, 25 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. The Model Validity Problem
  3. State-of-the-art Vehicle Models
  4. Four-wheel Model
  5. Bicycle Model
  6. Tire Models
  7. Pacejka Model
  8. Dugoff Model
  9. Linear Model
  10. Learned Vehicle Models
  11. Real Vehicle Definition and Data Collection
  12. Real Vehicle Behavior Analysis
  13. Comparison procedure
  14. Comparison Results
  15. Error visualization
  16. ...and 8 more sections

Figures (25)

  • Figure 1: The used experimental vehicle.
  • Figure 2: Four-wheel vehicle model.
  • Figure 3: The dynamic bicycle model.
  • Figure 4: A comparison plot between the different tire models showing the behavior difference with higher slips. (For representation purposes $B=10$, $C=2.2$, $D=2500$, $E=1$, $F_z=3000 N$)
  • Figure 5: Experimental sensor setup.
  • ...and 20 more figures