Latest Version of the Paper

The paper is currently under review at the Conference on Robot Learning (CoRL 2022).