# Information Theoretic Model Predictive Control: Theory and Applications to Autonomous Driving

**Authors:** Grady Williams, Paul Drews, Brian Goldfain, James M. Rehg, Evangelos A. Theodorou  
**arXiv:** [1707.02342](https://arxiv.org/abs/1707.02342)  
**Date:** 2017-07-07  
**Summary page:** https://www.summarizepaper.com/en/paper/1707.02342v1/

## Key points

- The paper presents a novel approach to stochastic optimal control problems using information theory.
- The authors propose a mathematical method for deriving general sampling-based optimization schemes.
- The method is applied to develop a sampling-based model predictive control (MPC) algorithm, referred to as IT-MPC.
- IT-MPC is evaluated on a dirt test track and compared with the cross-entropy method in terms of aggressive driving maneuvers.
- Results show that IT-MPC outperforms the cross-entropy method.
- The paper provides an in-depth analysis of the approach, including theoretical foundations and practical applications.
- IT-MPC is designed to handle uncertainties in autonomous driving scenarios, making it suitable for real-world applications.
- Overall, the paper contributes to the field of autonomous driving by introducing an innovative information theoretic approach to stochastic optimal control problems.

## Summary

The paper titled "Information Theoretic Model Predictive Control: Theory and Applications to Autonomous Driving" presents a novel approach to stochastic optimal control problems using information theory. The authors propose a mathematical method that allows for the derivation of general sampling-based optimization schemes. This method is then applied to develop a sampling-based model predictive control (MPC) algorithm. The main focus of this research is on applying the proposed information theoretic MPC scheme, referred to as IT-MPC, to the task of aggressive autonomous driving around a dirt test track. To evaluate the effectiveness of IT-MPC, the authors conduct experiments on a dirt test track and compare its performance with that of the cross-entropy method, which is widely used in model predictive control. The results show that IT-MPC outperforms the cross-entropy method in terms of aggressive driving maneuvers. The authors provide an in-depth analysis of their approach, including theoretical foundations and practical applications. They demonstrate how information theory can be leveraged to address stochastic optimal control problems and derive efficient optimization schemes. The IT-MPC algorithm is designed to handle uncertainties inherent in autonomous driving scenarios, making it suitable for real-world applications. Overall, this paper contributes to the field of autonomous driving by introducing an innovative information theoretic approach to stochastic optimal control problems. The proposed IT-MPC algorithm demonstrates promising results in enabling aggressive autonomous driving while considering uncertainties and optimizing control actions based on sampled data.