Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Meta-Learning of Vehicle Yaw Rate Dynamics via Conditional Neural Processes

Lars Ullrich, Andreas Völz, Knut Graichen

TL;DR

This work tackles yaw rate dynamics for autonomous-vehicle planning, where traditional physical models either oversimplify or incur high computational costs under varying conditions. It proposes a meta-learning approach using Conditional Neural Processes to learn yaw rate dynamics from diverse driving scenarios, yielding fast, uncertainty-aware predictions without online adaptation. Across friction, mass, scenarios, and vehicle transfer tests in CarMaker-based simulations, the CNP shows robust performance and transferability, approximating or exceeding more complex physics while providing uncertainty estimates. The study suggests that CNP-based yaw rate modeling can enable safer, more scalable trajectory planning with reduced model-order complexity, and highlights avenues for future enhancement and safety considerations.

Abstract

Trajectory planners of autonomous vehicles usually rely on physical models to predict the vehicle behavior. However, despite their suitability, physical models have some shortcomings. On the one hand, simple models suffer from larger model errors and more restrictive assumptions. On the other hand, complex models are computationally more demanding and depend on environmental and operational parameters. In each case, the drawbacks can be associated to a certain degree to the physical modeling of the yaw rate dynamics. Therefore, this paper investigates the yaw rate prediction based on conditional neural processes (CNP), a data-driven meta-learning approach, to simultaneously achieve low errors, adequate complexity and robustness to varying parameters. Thus, physical models can be enhanced in a targeted manner to provide accurate and computationally efficient predictions to enable safe planning in autonomous vehicles. High fidelity simulations for a variety of driving scenarios and different types of cars show that CNP makes it possible to employ and transfer knowledge about the yaw rate based on current driving dynamics in a human-like manner, yielding robustness against changing environmental and operational conditions.

Robust Meta-Learning of Vehicle Yaw Rate Dynamics via Conditional Neural Processes

TL;DR

This work tackles yaw rate dynamics for autonomous-vehicle planning, where traditional physical models either oversimplify or incur high computational costs under varying conditions. It proposes a meta-learning approach using Conditional Neural Processes to learn yaw rate dynamics from diverse driving scenarios, yielding fast, uncertainty-aware predictions without online adaptation. Across friction, mass, scenarios, and vehicle transfer tests in CarMaker-based simulations, the CNP shows robust performance and transferability, approximating or exceeding more complex physics while providing uncertainty estimates. The study suggests that CNP-based yaw rate modeling can enable safer, more scalable trajectory planning with reduced model-order complexity, and highlights avenues for future enhancement and safety considerations.

Abstract

Trajectory planners of autonomous vehicles usually rely on physical models to predict the vehicle behavior. However, despite their suitability, physical models have some shortcomings. On the one hand, simple models suffer from larger model errors and more restrictive assumptions. On the other hand, complex models are computationally more demanding and depend on environmental and operational parameters. In each case, the drawbacks can be associated to a certain degree to the physical modeling of the yaw rate dynamics. Therefore, this paper investigates the yaw rate prediction based on conditional neural processes (CNP), a data-driven meta-learning approach, to simultaneously achieve low errors, adequate complexity and robustness to varying parameters. Thus, physical models can be enhanced in a targeted manner to provide accurate and computationally efficient predictions to enable safe planning in autonomous vehicles. High fidelity simulations for a variety of driving scenarios and different types of cars show that CNP makes it possible to employ and transfer knowledge about the yaw rate based on current driving dynamics in a human-like manner, yielding robustness against changing environmental and operational conditions.
Paper Structure (14 sections, 14 equations, 5 figures, 5 tables)

This paper contains 14 sections, 14 equations, 5 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. PHYSICAL MODELS
  3. Kinematic Single-Track Model (KST)
  4. Dynamic Single-Track Model (DST)
  5. Single-Track Drift Model (STD)
  6. CONDITIONAL NEURAL PROCESSES (CNP)
  7. CNP Framework
  8. CNP based Yaw Rate Prediction
  9. EVALUATION RESULTS
  10. Friction variation
  11. Mass Variation
  12. Scenario variation
  13. Vehicle variation
  14. CONCLUSIONS AND FUTURE WORKS

Figures (5)

  • Figure 1: CNP for yaw rate predictions.
  • Figure 2: Evaluation under changed friction coefficients.
  • Figure 3: Evaluation of prediction under varied mass. Results separated according to different road (friction) conditions.
  • Figure 4: Prediction evaluation under varied scenarios. Results separated according to different road (friction) conditions.
  • Figure 5: Qualitative trajectory evaluation over (a) dry, (b) wet and (c) icy roads on the Hockenheim racetrack. The vertical dashed line separates the context set on the left from the predictions on the right. The time horizon in (c) is shorter since, given the same driving behavior, the vehicle leaves the road under icy conditions.