Risk-Sensitive Model Predictive Control

TL;DR

This paper introduces a risk-sensitive model predictive control approach that accounts for uncertain noise, providing a heuristic policy with performance bounds, and demonstrates its effectiveness in reducing risk in a battery control example.

Contribution

It extends standard MPC to incorporate risk sensitivity using a convex approximation and provides a performance bound for the proposed heuristic policy.

Findings

Reduces risk of bad outcomes compared to certainty equivalent control.

Maintains similar average performance while improving risk management.

Provides a convex formulation for risk-seeking cases and an approximate solution for risk-averse cases.

Abstract

We present a heuristic policy and performance bound for risk-sensitive convex stochastic control that generalizes linear-exponential-quadratic regulator (LEQR) theory. Our heuristic policy extends standard, risk-neutral model predictive control (MPC); however, instead of ignoring uncertain noise terms, our policy assumes these noise terms turns out either favorably or unfavorably, depending on a risk aversion parameter. In the risk-seeking case, this modified planning problem is convex. In the risk-averse case, it requires minimizing a difference of convex functions, which is done (approximately) using the convex-concave procedure. In both cases, we obtain a lower bound on the optimal cost as a by-product of solving the planning problem. We give a numerical example of controlling a battery to power an uncertain load, and show that our policy reduces the risk of a very bad outcome (as compared with standard certainty equivalent control) with negligible impact on the the average performance.