# A Comparison of Policy Search in Joint Space and Cartesian Space for Refinement of Skills

## TL;DR

This paper compares policy search for robot skill refinement in joint and Cartesian spaces, proposing a new inverse kinematic solver to improve Cartesian space refinement efficiency.

## Contribution

It introduces a configurable approximate inverse kinematic solver and empirically compares Cartesian and joint space policy search methods for skill refinement.

## Key findings

- The new inverse kinematic solver accelerates Cartesian space refinement.
- Empirical results show differences in efficiency between joint and Cartesian space methods.
- Refinement in Cartesian space can be improved with the proposed solver.

## Abstract

Imitation learning is a way to teach robots skills that are demonstrated by humans. Transfering skills between these different kinematic structures seems to be straightforward in Cartesian space. Because of the correspondence problem, however, the result will most likely not be identical. This is why refinement is required, for example, by policy search. Policy search in Cartesian space is prone to reachability problems when using conventional inverse kinematic solvers. We propose a configurable approximate inverse kinematic solver and show that it can accelerate the refinement process considerably. We also compare empirically refinement in Cartesian space and refinement in joint space.