pi0.5 CubeStack โ€” local-only c0 (harry, Green/Orange)

From-scratch local baseline: harry's 2 tasks ONLY (Green_on_Orange, Orange_on_Green), 10000 steps, own 2-repo normalizer. Blind to Blue -> expected to FAIL the 4 Blue tasks. Floor of the hierarchy.

lerobot-native export (portable tokenizer, real 6-D action/state; --policy.compile_model=false). SO-101 CubeStack, cameras base_0_rgb/left_wrist_0_rgb. Trained on Isambard-AI (GH200) with decent-vla. Part of the 3-client pi0.5 CubeStack federated study (color-pair non-IID: c0=harry Green/Orange, c1=zhekai Green/Blue, c2=kevin Orange/Blue; each client blind to the 3rd color).

Downloads last month
-
Safetensors
Model size
4B params
Tensor type
F32
ยท
Video Preview
loading

Model tree for DecentVLA/pi05_cubestack_local3_c0

Finetuned
(146)
this model