Adaptive Risk Sensitive Model Predictive Control with Stochastic Search

TL;DR

This paper introduces a stochastic search-based framework for optimizing Conditional Value-at-Risk in dynamical systems, effectively managing uncertainties and outperforming existing risk-sensitive methods in robotics applications.

Contribution

It presents a novel, general approach for risk-sensitive control using stochastic search, capable of handling various uncertainties in dynamical systems.

Findings

Outperforms risk-sensitive distributional reinforcement learning on benchmark systems

Successfully applies to adaptive risk-sensitive control in robotics

Demonstrates effectiveness in simulation for pendulum, cartpole, and quadcopter

Abstract

We present a general framework for optimizing the Conditional Value-at-Risk for dynamical systems using stochastic search. The framework is capable of handling the uncertainty from the initial condition, stochastic dynamics, and uncertain parameters in the model. The algorithm is compared against a risk-sensitive distributional reinforcement learning framework and demonstrates outperformance on a pendulum and cartpole with stochastic dynamics. We also showcase the applicability of the framework to robotics as an adaptive risk-sensitive controller by optimizing with respect to the fully nonlinear belief provided by a particle filter on a pendulum, cartpole, and quadcopter in simulation.