We're doing exactly this to teleoperate humanoid robots on high-latency networks! Paper: https://arxiv.org/abs/2107.01281 https://arxiv.org/abs/2107.01281 Vid

We're doing exactly this to teleoperate humanoid robots on high-latency networks! Paper: https://arxiv.org/abs/2107.01281 https://arxiv.org/abs/2107.01281 Video: https://www.youtube.com/watch?v=N3u4ot3aIyQ https://www.youtube.com/watch?v=N3u4ot3aIyQ "We introduce a system in which a humanoid robot executes commands before it actually receives them, so that the visual feedback appears to be synchronized to the operator, whereas the robot executed the commands in the past. To do so, the robot continuously predicts future commands by querying a machine learning model that is trained on past trajectories and conditioned on the last received commands. In our experiments, an operator was able to successfully control a humanoid robot (32 degrees of freedom) with stochastic delays up to 2 seconds in several whole-body manipulation tasks, including reaching different targets, picking up, and placing a box at distinct locations."