The muscle body: muscles instead of joint motors

Every fighter so far has been driven by motors in its joints: the policy picks a target angle for each joint and a servo gets it there, so an action of zero already tracks the reference motion. A real body has muscles, which only pull, lag behind the command, and are stiff only when opposing muscles both work. On 27 October I asked how much longer training would take with muscles, whether it works at all, and how much of the existing work could be reused.

The first stage keeps everything else the same: the same skeleton, masses, contacts and reference motions, with a pair of muscles across each joint in place of its motor (86 in all). That makes the comparison with the servo body direct. Real muscle paths and muscles across two joints come later, and only if this stage works. The budget for these first stages is about $15.

The runs

run	what	status
mbase	the servo body learning the stance from scratch, two seeds, on the same RunPod RTX 4090 and settings as the muscle runs: the baseline to beat	training
muscle1s0, muscle1s1	the muscle body learning the stance from scratch, seeds 0 and 1, up to 3 h each	training

Before them, a short check on the same kind of RunPod machine confirmed the GPU simulation of the muscle body matches the CPU one. The results go here when the runs end.

Every run

From the run records in kakuto-ml: each run's machines, their billed hours and cost (EC2 runs are estimates from the run logs), plus the tests and renders this log covers.

• GPU against CPU parity check of the muscle body on a RunPod RTX 4090, 27 October: 1 machines, 0.1 h, $0.08

← All the work logs