Reacher observation appears to be mislabled

[jackhanslope]

In the docstring for the Reacher environment, two of the elements of the observation are the angular velocities of the two arms.

Inside _get_obs it seems to me that the two elements that are the "angular velocities" in the observation are actually the x and y coordinate of the linear velocity of the fingertip (relative to the world).

Relevant snippet (some lines omitted) from Reacher._get_obs

arm_qps = jp.take(qp, jp.array(self._arm_idx))
tip_pos, tip_vel = arm_qps.to_world(jp.array([0.11, 0., 0.]))

qvel = [tip_vel[:2]]

[btaba]

Hi @jackhanslope, @erikfrey or @cdfreeman-google may be able to help here