Reproduce “The Hybrid Extended Bicycle: A Simple Model for High Dynamic Vehicle Trajectory Planning” on OpenResearch

About this paper

Agapius Bou Ghosn, Philip Polack, Arnaud de La Fortelle

The Hybrid Extended Bicycle Model (HEBM) combines a kinematic bicycle model with recurrent neural networks to predict wheel slip angles, enabling accurate vehicle trajectory planning under high-dynamic conditions. Integrated into a Model Predictive Path Integral (MPPI) planner, the HEBM demonstrated superior path adherence and velocity control compared to a kinematic bicycle model, achieving lateral accelerations up to 0.76g in simulations.

Key insights

• Augmenting a simple physics-based vehicle model with a machine learning component for slip angle prediction can effectively bridge the gap between computational efficiency and dynamic accuracy.
• The learned slip angle prediction allows a relatively simple hybrid model to capture complex, nonlinear tire dynamics, enabling accurate trajectory planning even at the vehicle's handling limits.
• Model Predictive Path Integral (MPPI) is well-suited for planning with hybrid models containing neural network components, as its derivative-free nature naturally handles the implicit complexities without requiring analytical gradients.

Summary by alphaXiv.