Inverse optimal control from incomplete trajectory observations

This article develops a methodology that enables learning an objective function of optimal control system from incomplete trajectory observations. The is assumed to be weighted sum features (or basis functions) with unknown weights, and the observed data segment states inputs. proposed technique introduces concept recovery matrix establish relationship between any available weights given candidate features. rank indicates whether subset relevant can found among corresponding learned data. obtained iteratively its non-decreasing property shows additional observations may contribute learning. Based on matrix, method for using learn selected established, incremental inverse algorithm developed by automatically finding minimal required observation. effectiveness demonstrated linear quadratic regulator simulated robot manipulator.