gin-sock-s3

Counter with one re-searched seed, 4.779 s in the local x86 replica [4.635, 5.095, 4.670, 4.680, 4.815]. Seeds 1, 2, 4 and 5 replay rumezhak's microduck-sock-heist-composite-v10 unchanged (run 01M7N0C6V63JMP4M9FM2G1; its seeds 1, 2, 3, 5 are HlibPlakhtii's nov17-v3 on jrubiosainz's counter-s1s5, its seed 4 is rumezhak's search from lphatnguyen's v5). Seed 3 is own work on that trajectory: an elitist tail search over the last 60 calls with a prefix cache (only the tail is simulated) and a continuous crossing time, 4.675 -> 4.670 s. The prefix cache and tail search follow the method HlibPlakhtii and rumezhak describe in their cards; the code is written from scratch. Fallback net: own. Simulation only. Credits: rumezhak, HlibPlakhtii, jrubiosainz, lphatnguyen.

A perpetual policy for the microduck (61-D observation, 14 actions, 50 Hz). Runs until told otherwise โ€” a gait, loaded into a policy slot.

Run it on a robot

sudo robotctl policy load <slot> GinDuck/microduck-gin-sock-s3

The observation normalizer is baked into policy.onnx; feed raw observations. manifest.json follows schema 2 of the microduck policy manifest (docs/policy-manifest.md in the daemon repo).

Training

  • repo: pollen-robotics/microduck_rl
  • branch: talent-3
  • commit: debcfcae0
Downloads last month
19
Video Preview
loading

Model tree for GinDuck/microduck-gin-sock-s3

Finetunes
1 model