Statistically Consistent Inverse Optimal Control for Linear-Quadratic Tracking with Random Time Horizon

TL;DR

This paper introduces a statistically consistent inverse optimal control algorithm for linear-quadratic tracking with random time horizons, validated on simulated and real human locomotion data, useful for personalized robotics.

Contribution

It develops a convex optimization-based IOC algorithm with proven statistical consistency for complex tracking problems with random horizons.

Findings

Algorithm achieves statistical consistency in synthetic data.

Real data experiments show accurate prediction of human locomotion.

Identified models can inform personalized rehabilitation control.

Abstract

The goal of Inverse Optimal Control (IOC) is to identify the underlying objective function based on observed optimal trajectories. It provides a powerful framework to model expert's behavior, and a data-driven way to design an objective function so that the induced optimal control is adapted to a contextual environment. In this paper, we design an IOC algorithm for linear-quadratic tracking problems with random time horizon, and prove the statistical consistency of the algorithm. More specifically, the proposed estimator is the solution to a convex optimization problem, which means that the estimator does not suffer from local minima. This enables the proven statistical consistency to actually be achieved in practice. The algorithm is also verified on simulated data as well as data from a real world experiment, both in the setting of identifying the objective function of human tracking locomotion. The statistical consistency is illustrated on the synthetic data set, and the experimental results on the real data shows that we can get a good prediction on human tracking locomotion based on estimating the objective function. It shows that the theory and the model have a good performance in real practice. Moreover, the identified model can be used as a control target in personalized rehabilitation robot controller design, since the identified objective function describes personal habit and preferences.