A Data-Driven Aggressive Autonomous Racing Framework Utilizing Local Trajectory Planning with Velocity Prediction

TL;DR

This work tackles the challenge of local trajectory planning for aggressive autonomous racing, especially around sharp corners, by introducing Velocity Prediction MPCC (VPMPCC) that explicitly predicts velocity via a Reference Velocity Profile (RVP). The VPMPCC is trained efficiently with Bayesian Optimization using an Objective Function Adapted to Racing (OFR), which balances lap-time improvement and safety, and enables safe transfer from simulation to real-world vehicles. The framework, VPBO-RF, integrates a high-fidelity dynamics-based trajectory filter and uses an EI-based BO loop to learn VPMPCC parameters, achieving faster convergence and near-limit performance. Real-world tests on a scaled F1TENTH vehicle show VPMPCC achieving mean projected velocity at 93.18% of handling limits with low computation time, demonstrating practical viability for high-speed autonomous racing.

Abstract

The development of autonomous driving has boosted the research on autonomous racing. However, existing local trajectory planning methods have difficulty planning trajectories with optimal velocity profiles at racetracks with sharp corners, thus weakening the performance of autonomous racing. To address this problem, we propose a local trajectory planning method that integrates Velocity Prediction based on Model Predictive Contouring Control (VPMPCC). The optimal parameters of VPMPCC are learned through Bayesian Optimization (BO) based on a proposed novel Objective Function adapted to Racing (OFR). Specifically, VPMPCC achieves velocity prediction by encoding the racetrack as a reference velocity profile and incorporating it into the optimization problem. This method optimizes the velocity profile of local trajectories, especially at corners with significant curvature. The proposed OFR balances racing performance with vehicle safety, ensuring safe and efficient BO training. In the simulation, the number of training iterations for OFR-based BO is reduced by 42.86% compared to the state-of-the-art method. The optimal simulation-trained parameters are then applied to a real-world F1TENTH vehicle without retraining. During prolonged racing on a custom-built racetrack featuring significant sharp corners, the mean projected velocity of VPMPCC reaches 93.18% of the vehicle's handling limits. The released code is available at https://github.com/zhouhengli/VPMPCC.

Figures (7)

• Figure 1: Comparison of trajectories between the proposed Velocity Prediction MPCC (VPMPCC) and standard MPCC through a high-curvature corner using a self-built F1TENTH vehicle. The trajectories are captured through long exposures. VPMPCC is able to plan optimal cornering trajectories at higher velocities, resulting in shorter cornering time to reduce lap time.
• Figure 2: Schematic diagram illustrating the calculation of contouring and lag errors in conventional MPCC. The proposed VPMPCC method incorporates the vehicle's longitudinal velocity as an independent decision variable.
• Figure 3: Schematic diagram of the VPBO-RF. The brown arrows indicate the online evaluation process of parameter $\bm{\theta}_i$. The light brown arrows indicate the next evaluation parameter $\bm{\theta}_{i+1}$ chosen by the acquisition function. The closed-loop training convergence means that the OFR value no longer decreases.
• Figure 4: The grid map created from real racetracks determines boundaries for VPMPCC and is used for localization during real vehicle racing.
• Figure 5: The different colored boxes represent the optimal lap time obtained from training in Fig. \ref{['fig:J_train']}. The curves indicate the current best points during training. The proposed Objective Function adapted to Racing enables BO to converge rapidly, and its optimal lap time outperforms the baseline. Ablation experiments demonstrate the effectiveness of different components of OFR for improving the training efficiency and safety of optimal trajectories.