Boston Dynamics gives Atlas hands that lift a fridge
Atlas got 13-degree-of-freedom hands that are directly actuated, not tendon driven, and Chris Paxton says they can still carry a 100+ lb fridge. His read on the tendon-driven competition: it may be nerd-sniping.
Shown this week
Boston Dynamics gives Atlas 13-DOF hands built for sim-to-real RL
Boston Dynamics announced new Atlas hands with 13 degrees of freedom, directly actuated and designed for high fidelity simulation so policies can be trained in sim and run on the robot.
I'm cautiously optimistic about many of the tendon-drive hands coming onto the market (proception is so cool!) but sometimes I worry it's a bit of nerd-sniping.
Paxton says Mind-1's fast humanlike videos are probably not teleop
MindOn posted Mind-1, pitched as physical AI running at human speed, and Chris Paxton says the videos are consistently mind blowing: very fast, smooth, humanlike motion. He says he is fairly sure it is UMI-style work (training from handheld human demonstrations), not teleop data, and wants to know more about the stack.
Fairly sure this is UMI-style work, not teleop data.
Teleop and other arguments
Researchers fight over whether RL should replace IK entirely
Qiayuan Liao argued that any online optimization for low-level control, including inverse kinematics (solving joint angles for a target position), should be abandoned in favor of an RL policy, which he says tracks more precisely and handles unknown disturbances, especially on low-cost hardware.
The future clearly is not in model based control, even for wheeled robots