TY - JOUR
T1 - RAT iLQR
T2 - A Risk Auto-Tuning Controller to Optimally Account for Stochastic Model Mismatch
AU - Nishimura, Haruki
AU - Mehr, Negar
AU - Gaidon, Adrien
AU - Schwager, Mac
N1 - Manuscript received October 15, 2020; accepted December 12, 2020. Date of publication January 1, 2021; date of current version January 29, 2021. This letter was recommended for publication by Associate Editor A. Quattrini Li and Editor L. Pallottino upon evaluation of the reviewers’ comments. This work was supported in part by ONR under Grant N00014-18-1-2830, NSF under Grant NRI 1830402, in part by JASSO fellowship, and in part by Masason Foundation fellowship. (Corresponding author: Haruki Nishimura.) Haruki Nishimura and Mac Schwager are with the Department of Aeronautics and Astronautics, Stanford University, Stanford, CA 94305 USA (e-mail: [email protected]; [email protected]).
PY - 2021/4
Y1 - 2021/4
N2 - Successful robotic operation in stochastic environments relies on accurate characterization of the underlying probability distributions, yet this is often imperfect due to limited knowledge. This work presents a control algorithm that is capable of handling such distributional mismatches. Specifically, we propose a novel nonlinear MPC for distributionally robust control, which plans locally optimal feedback policies against a worst-case distribution within a given KL divergence bound from a Gaussian distribution. Leveraging mathematical equivalence between distributionally robust control and risk-sensitive optimal control, our framework also provides an algorithm to dynamically adjust the risk-sensitivity level online for risk-sensitive control. The benefits of the distributional robustness as well as the automatic risk-sensitivity adjustment are demonstrated in a dynamic collision avoidance scenario where the predictive distribution of human motion is erroneous.
AB - Successful robotic operation in stochastic environments relies on accurate characterization of the underlying probability distributions, yet this is often imperfect due to limited knowledge. This work presents a control algorithm that is capable of handling such distributional mismatches. Specifically, we propose a novel nonlinear MPC for distributionally robust control, which plans locally optimal feedback policies against a worst-case distribution within a given KL divergence bound from a Gaussian distribution. Leveraging mathematical equivalence between distributionally robust control and risk-sensitive optimal control, our framework also provides an algorithm to dynamically adjust the risk-sensitivity level online for risk-sensitive control. The benefits of the distributional robustness as well as the automatic risk-sensitivity adjustment are demonstrated in a dynamic collision avoidance scenario where the predictive distribution of human motion is erroneous.
KW - Collision avoidance
KW - optimization and optimal control
KW - robust/adaptive control
UR - https://www.scopus.com/pages/publications/85099098417
UR - https://www.scopus.com/pages/publications/85099098417#tab=citedBy
U2 - 10.1109/LRA.2020.3048660
DO - 10.1109/LRA.2020.3048660
M3 - Article
AN - SCOPUS:85099098417
SN - 2377-3766
VL - 6
SP - 763
EP - 770
JO - IEEE Robotics and Automation Letters
JF - IEEE Robotics and Automation Letters
IS - 2
M1 - 9312440
ER -