Imitation learning systems often train on teleoperated demonstrations, and researchers increasingly rely on novice operators to collect them. But non-experts provide fewer demonstrations and are less consistent across trials. A new preprint from arXiv argues that the interface itself matters: operators act on what they are shown, so the visual rendering of the robot can influence the demonstrations that go into the policy.
The paper reports a user study of keypoint hand visual representation for teleoperated demonstration. The central question is whether showing the robot hand as keypoints—rather than some other rendering—changes how novices move and what the resulting policy learns. The abstract makes clear that the rendering affects operator behaviour, and the title states that what novice operators see changes what the policy learns.
Because only one source was available for this digest, there are no independent results to compare or contrast. The full experimental findings are not included in the abstract, so this summary is limited to the study's framing and stated claim.