Project 16: Background Review

A major limitation in applying existing methods for inverse optimal control (IOC) to robotic examples is the high dimensionality of these examples. I am interested in addressing this by looking in to recent work by Levine et al [2]. This paper an interesting inverse optimal control paper that goes into an method for inverse optimal control on continuous valued data rather than discrete valued data, as many previous papers assume (e.g. [4]). This work also has the added advantage of using locally optimal, rather than globally optimal examples: this means that if the human is limited by clumsy controls the system might be able to learn a better approach.