Physical AI: The Robotic Rise and Eventual Fall of the Stochastic Puppet | by Brad Porter | Medium

For the past couple of years, the number of exciting robot videos has increased dramatically. Many of these are showing tasks we’ve never robots do before (e.g. folding laundry, talking to guests at a party). By and large these videos are enabled by two currently popular techniques: teleoperation and imitation learning. Teleoperation is the simpler of these to explain. Teleoperation is remotely controlling a robot, much like a remote-control car or RC airplane. A human uses a controller or some device or suit or maybe motion capture, and those commands are relayed to the robot and the robot performs those commands directly. If this sounds like puppeteering or Wizard of Oz demonstrations, that’s because that’s what it is. Imitation learning is an extension of teleoperation. In imitation learning, the human performs the same task over and over again and the commands the human operators made are saved. By saving the commands, you can replay the same actions over and over. You’ve just memo

For the past couple of years, the number of exciting robot videos has increased dramatically. Many of these are showing tasks we’ve never robots do before (e.g. folding laundry, talking to guests at a party). By and large these videos are enabled by two currently popular techniques: teleoperation and imitation learning. Teleoperation is the simpler of these to explain. Teleoperation is remotely controlling a robot, much like a remote-control car or RC airplane. A human uses a controller or some device or suit or maybe motion capture, and those commands are relayed to the robot and the robot pe