Demo: Deadline-Aware Real-Time Planning on F1TENTH

Abstract

We demonstrate a complete, real-time autonomous racing stack that deploys Deadline-Aware Monte Carlo Tree Search (DA-MCTS) on physical F1TENTH hardware powered by an NVIDIA Jetson Orin Nano. DA-MCTS adapts the risk profile of its sampled actions to the compute budget available at runtime, letting a sampling-based planner meet a hard control deadline on a resource-constrained processor without retuning. On F1TENTH hardware, DA-MCTS computes a racing line online and attains lap times competitive with pure-pursuit and Stanley baselines that require a precomputed reference line. Demo video link: https://youtu.be/vgiwadLAvKs and source code link: https://github.com/scope-lab-vu/RTCSA_Demo_ws.