You can deploy this model and start earning money today!

chennana1028/
bc-mae-openarm-aist-exp18-nolora-chunk16

Name: bc-mae-openarm-aist-exp18-nolora-chunk16
Availability: InStock

Views

No views yet

BC MAE + MLP (NoLoRA encoder, action chunk H=16)

Frozen MAE (OpenArm AIST exp18 NoLoRA) + BC MLP with action chunking (H=16). Data: SBInt OpenArm pnp image_filtered.hdf5 (187 demos, ~43k frames).

Sister model (LoRA encoder): chennana1028/bc-mae-openarm-aist-exp18-lora-chunk16.

Checkpoints (saved every 20)

File	Epoch	val_loss	train_loss
`epoch=0040.ckpt`	40	0.0132	0.0072
`epoch=0080.ckpt`	80	0.0136	0.0049
`epoch=0100.ckpt`	100	0.0168	0.0043
`epoch=0120.ckpt`	120	0.0140	0.0030
`epoch=0140.ckpt`	140	0.0128	0.0028
`epoch=0200.ckpt`	200	0.0142	0.0016
`epoch=0480.ckpt`	480	0.0129	0.0006
`epoch=0499.ckpt`	499	0.0142	0.0008
`best_val.ckpt`	140	~0.0128	(alias of ep140)
`latest.ckpt`	499	~0.0142	alias of final

Load

python

from policy import load_policy
policy = load_policy("best_val.ckpt", device="cuda:0")
out = policy.predict_action(obs)["action"]  # (B, 16, 8)

Obs: agentview_image, robot0_eye_in_hand_image (B,3,224,224) float [0,1], robot0_joint_qpos (B,8). Deploy: execute first k steps of the chunk @ ~30 Hz, then replan (do not only send action[0] forever).

Train note

Only the MLP head is trained on a frozen MAE feature cache (8×GPU, bs=64). 500 epochs ≈ 15 min wall — that is expected, not a bug. MAE encoder: mae_openarm_aist_exp18_nolora/encoder.pth.

chennana1028/bc-mae-openarm-aist-exp18-nolora-chunk16

BC MAE + MLP (NoLoRA encoder, action chunk H=16)

Recommended

Checkpoints (saved every 20)

Load

Train note

chennana1028/
bc-mae-openarm-aist-exp18-nolora-chunk16