Reproduce “Information Theoretic Model Predictive Control: Theory and Applications to Autonomous Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Grady Williams, Paul Drews, Brian Goldfain, James M. Rehg, Evangelos A. Theodorou
This paper presents Information Theoretic Model Predictive Control (IT-MPC), a sampling-based optimal control algorithm that reinterprets stochastic optimal control through an information-theoretic framework. The method enables real-time high-performance autonomous driving, successfully demonstrated on a 1:5 scale AutoRally vehicle where it robustly executed aggressive maneuvers near and beyond the vehicle's friction limits.
Key insights
- Stochastic optimal control can be reinterpreted as minimizing the Kullback-Leibler (KL) divergence between a controlled distribution and an optimal distribution of trajectories.
- The optimal control sequence can be computed as a weighted average of sampled trajectories, where weights are exponentially related to trajectory costs (Boltzmann-like distribution).
- Practical enhancements like cost shifting, temperature parameter decoupling (alpha), and simple constraint clamping enable real-time performance and numerical stability for the sampling-based IT-MPC algorithm.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.