pollen-robotics/microduck-training

Training checkpoints of pollen-robotics/microduck_rl policies (full rsl_rl .pt: weights, normalizers, optimizer, curriculum step counter), to resume or warm-start training from. The deployable ONNX policies live in their own repos; each bundle below records which ONNX it was exported to.

uv run train <task> --init-checkpoint hf://pollen-robotics/microduck-training/bundles/<experiment>/<run>_<iter>

bundle	task	exported to	code
ball_kick_left/6bamlid8_2500	Mjlab-BallKick-Flat-MicroDuck	pollen-robotics/microduck-policies/ball_kick_left.onnx	8c7665695
ball_kick_right/wy2x3p87_2000	Mjlab-BallKick-Flat-MicroDuck	pollen-robotics/microduck-policies/ball_kick_right.onnx	8c7665695
ground_pick/4ly1pcxj_2750	Mjlab-GroundPick-Flat-MicroDuck	pollen-robotics/microduck-policies/alpha_ground_pick.onnx	f93771b30
microduck_roulade/blrqbcit_750	Mjlab-Roulade-Flat-MicroDuck	pollen-robotics/microduck-policies/roulade.onnx	?
microduck_sitstand/vfhxgrum_2250	Mjlab-SitStand-Flat-MicroDuck	pollen-robotics/microduck-policies/alpha_sitstand.onnx	77ffadf43
microduck_stand/69u48n8l_9750	Mjlab-StandUp-Flat-MicroDuck	pollen-robotics/microduck-policies/alpha_stand.onnx	85105f4b1
roller_crouch/92d20zda_500	Mjlab-RollerCrouch-Flat-MicroDuck	pollen-robotics/microduck-policies/roller_crouch.onnx	1ae24ae8f
velocity2/441tzs6d_3750	Mjlab-Velocity2-Rough-Backlash-MicroDuck	pollen-robotics/microduck-policies/alpha_walking.onnx	2f1f76bc6
velstand/0045i2si_2749	Mjlab-VelStand-Rough-Backlash-MicroDuck	pollen-robotics/microduck-policies/velstand.onnx	375077676

Written by uv run share-training (see docs/training_checkpoints.md in the code repo).