Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Process-Based Reactive Controller for Autonomous Racing

Devin Hunter, Chinwendu Enyioha

TL;DR

This work tackles safe, high-speed autonomous racing by learning a follow-the-gap policy using an Attentive Neural Process (AttNP) and its physics-informed variant (PI-AttNP). The PI-AttNP blends a data-driven predictor with an approximate FTG prior and a real-time Control Barrier Function (CBF) based safety filter, trained via a variational ELBO using Gaussian surrogates. Key contributions include the PI-AttNP architecture, the ELBO-based training regime, a speed heuristic, and a QP-based steering filter with explicit Lie-derivative safety constraints, enabling real-time constraint satisfaction. In simulated F1Tenth environments, PI-AttNP achieves faster convergence and lower collision rates (e.g., PI-AttNP with CBF has 0.2 collisions per run) while maintaining competitive lap times, highlighting practical viability for safety-critical autonomous racing.

Abstract

Attention-based neural architectures have become central to state-of-the-art methods in real-time nonlinear control. As these data-driven models continue to be integrated into increasingly safety-critical domains, ensuring statistically grounded and provably safe decision-making becomes essential. This paper introduces a novel reactive control framework for gap-based navigation using the Attentive Neural Process (AttNP) and a physics-informed extension, the PI-AttNP. Both models are evaluated in a simulated F1TENTH-style Ackermann steering racecar environment, chosen as a fast-paced proxy for safety-critical autonomous driving scenarios. The PI-AttNP augments the AttNP architecture with approximate model-based priors to inject physical inductive bias, enabling faster convergence and improved prediction accuracy suited for real-time control. To further ensure safety, we derive and implement a control barrier function (CBF)-based filtering mechanism that analytically enforces collision avoidance constraints. This CBF formulation is fully compatible with the learned AttNP controller and generalizes across a wide range of racing scenarios, providing a lightweight and certifiable safety layer. Our results demonstrate competitive closed-loop performance while ensuring real-time constraint satisfaction.

Neural Process-Based Reactive Controller for Autonomous Racing

TL;DR

This work tackles safe, high-speed autonomous racing by learning a follow-the-gap policy using an Attentive Neural Process (AttNP) and its physics-informed variant (PI-AttNP). The PI-AttNP blends a data-driven predictor with an approximate FTG prior and a real-time Control Barrier Function (CBF) based safety filter, trained via a variational ELBO using Gaussian surrogates. Key contributions include the PI-AttNP architecture, the ELBO-based training regime, a speed heuristic, and a QP-based steering filter with explicit Lie-derivative safety constraints, enabling real-time constraint satisfaction. In simulated F1Tenth environments, PI-AttNP achieves faster convergence and lower collision rates (e.g., PI-AttNP with CBF has 0.2 collisions per run) while maintaining competitive lap times, highlighting practical viability for safety-critical autonomous racing.

Abstract

Attention-based neural architectures have become central to state-of-the-art methods in real-time nonlinear control. As these data-driven models continue to be integrated into increasingly safety-critical domains, ensuring statistically grounded and provably safe decision-making becomes essential. This paper introduces a novel reactive control framework for gap-based navigation using the Attentive Neural Process (AttNP) and a physics-informed extension, the PI-AttNP. Both models are evaluated in a simulated F1TENTH-style Ackermann steering racecar environment, chosen as a fast-paced proxy for safety-critical autonomous driving scenarios. The PI-AttNP augments the AttNP architecture with approximate model-based priors to inject physical inductive bias, enabling faster convergence and improved prediction accuracy suited for real-time control. To further ensure safety, we derive and implement a control barrier function (CBF)-based filtering mechanism that analytically enforces collision avoidance constraints. This CBF formulation is fully compatible with the learned AttNP controller and generalizes across a wide range of racing scenarios, providing a lightweight and certifiable safety layer. Our results demonstrate competitive closed-loop performance while ensuring real-time constraint satisfaction.
Paper Structure (19 sections, 15 equations, 4 figures, 4 tables)

This paper contains 19 sections, 15 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction & Background
  2. Problem Statement
  3. Methodology
  4. Model Forward Pass Discussion
  5. Model Optimization Algorithm
  6. Speed Heuristic Utilized for FTG Algorithm
  7. Control Barrier Function (CBF) Formulation for Steering Safety
  8. Explicit Derivation of Vehicle Dynamics in Control-Affine Form
  9. Safety Function Definition
  10. CBF Constraint Derivation
  11. QP-Based Steering Filter
  12. Theoretical Guarantees
  13. Results & Discussion
  14. Approximate Control Prior for PI-AttNP
  15. Context/Target Formulation
  16. ...and 4 more sections

Figures (4)

  • Figure 1: Computational Graph of PI-AttNP's Forward Pass
  • Figure 2: Screenshot of simulated F1Tenth racing environment with ego vehicle. Note that all controllers used in racing environments were low-latency ($\sim200$ hz) control methods.
  • Figure 3: MAE convergence dynamics of the PI-AttNP, AttNP, and Res-MLP
  • Figure 4: Distribution error dynamics of the PI-AttNP, AttNP, and Res-MLP