SGTP: Sampling-based Game-Theoretic Planning for Real-Time Multi-Vehicle Autonomous Racing

Researchers introduced SGTP, a real-time game-theoretic planner for multi-vehicle autonomous racing.

The method samples control sequences and dynamics rollouts, scores trajectories with a competitive cost, and enforces safety constraints.

GPU acceleration enables fast computation, with simulations showing strong win and completion rates and low runtime.

The framework scales to 10 vehicles and comes with an open-source code release and benchmark for future research.