Dimios45/yam-pick-duster-200-ee-rollouts

YAM rollout log

Generated 2026-08-19 14:51.

A grasp means the measured gripper closed below 0.16 m and then lifted more than 5 cm while still closed. Closing in mid-air does not count.

data/pick200-ee-vrkit

2/4 grasps.

0/1 — ckpt_path=deploy_ema.ckpt policy=bspline control_freq=100.0 data_freq=10.0 origin_time_scale=10.0 num_inference_steps=8 predict_before_end=0.7 speed_up_times=1.0 max_steps=3000 device=cpu

episode	dur (s)	grasp	z@close	lift	stall %	speed (rad/s)
20260819T141734294652	29.3	no	—	—	nan	0.176

/home/yambox/bspline-policy/.venv/bin/python real_env/yam_teleop/rollout_local_policy.py --env yam-vrkit --policy bspline --ckpt-path /home/yambox/bspline-policy/ckpt_pick200_ee/deploy_ema.ckpt --diffusion-policy-dir /home/yambox/bspline-policy/diffusion_policy --device cpu --control-freq 100 --data-freq 10 --origin-time-scale 10 --num-inference-steps 8 --predict-before-end 0.7 --speed-up-times 1.0 --max-steps 3000 --num-episodes 10 --save --output-dir data/pick200-ee-vrkit

1/1 — ckpt_path=deploy_ema.ckpt policy=bspline control_freq=100.0 data_freq=10.0 origin_time_scale=10.0 num_inference_steps=16 predict_before_end=0.7 speed_up_times=1.4 max_steps=6000 device=cpu

episode	dur (s)	grasp	z@close	lift	stall %	speed (rad/s)
20260819T142121370460	21.4	yes	0.114	0.136	nan	0.541

/home/yambox/bspline-policy/.venv/bin/python real_env/yam_teleop/rollout_local_policy.py --env yam-vrkit --policy bspline --ckpt-path /home/yambox/bspline-policy/ckpt_pick200_ee/deploy_ema.ckpt --diffusion-policy-dir /home/yambox/bspline-policy/diffusion_policy --device cpu --control-freq 100 --data-freq 10 --origin-time-scale 10 --num-inference-steps 16 --predict-before-end 0.7 --speed-up-times 1.4 --max-steps 6000 --num-episodes 10 --save --output-dir data/pick200-ee-vrkit

1/2 — ckpt_path=deploy_ema.ckpt policy=bspline control_freq=100.0 data_freq=10.0 origin_time_scale=10.0 num_inference_steps=16 predict_before_end=0.7 speed_up_times=1.0 max_steps=6000 device=cpu

episode	dur (s)	grasp	z@close	lift	stall %	speed (rad/s)
20260819T142442684386	58.9	yes	0.112	0.132	nan	0.367
20260819T142551683883	58.9	no	0.119	0.001	nan	0.122

/home/yambox/bspline-policy/.venv/bin/python real_env/yam_teleop/rollout_local_policy.py --env yam-vrkit --policy bspline --ckpt-path /home/yambox/bspline-policy/ckpt_pick200_ee/deploy_ema.ckpt --diffusion-policy-dir /home/yambox/bspline-policy/diffusion_policy --device cpu --control-freq 100 --data-freq 10 --origin-time-scale 10 --num-inference-steps 16 --predict-before-end 0.7 --speed-up-times 1 --max-steps 6000 --num-episodes 10 --save --output-dir data/pick200-ee-vrkit

data/pick200-ee-vrkit_0.7x

3/5 grasps.

3/5 — ckpt_path=deploy_ema.ckpt policy=bspline control_freq=100.0 data_freq=10.0 origin_time_scale=10.0 num_inference_steps=16 predict_before_end=0.7 speed_up_times=0.7 max_steps=6000 device=cpu

episode	dur (s)	grasp	z@close	lift	stall %	speed (rad/s)
20260819T144123597533	58.9	yes	0.114	0.130	nan	0.276
20260819T144232550888	58.9	yes	0.108	0.139	nan	0.329
20260819T144341636823	58.9	no	—	—	nan	0.181
20260819T144450765826	58.9	yes	0.114	0.144	nan	0.277
20260819T144559646815	58.9	no	0.117	0.000	nan	0.138

/home/yambox/bspline-policy/.venv/bin/python real_env/yam_teleop/rollout_local_policy.py --env yam-vrkit --policy bspline --ckpt-path /home/yambox/bspline-policy/ckpt_pick200_ee/deploy_ema.ckpt --diffusion-policy-dir /home/yambox/bspline-policy/diffusion_policy --device cpu --control-freq 100 --data-freq 10 --origin-time-scale 10 --num-inference-steps 16 --predict-before-end 0.7 --speed-up-times 0.7 --max-steps 6000 --num-episodes 5 --save --output-dir data/pick200-ee-vrkit_0.7x

data/pick200-ee-vrkit_1.5x

4/5 grasps.

4/5 — ckpt_path=deploy_ema.ckpt policy=bspline control_freq=100.0 data_freq=10.0 origin_time_scale=10.0 num_inference_steps=16 predict_before_end=0.7 speed_up_times=1.5 max_steps=6000 device=cpu

episode	dur (s)	grasp	z@close	lift	stall %	speed (rad/s)
20260819T143034491520	21.6	yes	0.105	0.133	nan	0.739
20260819T143110125426	32.0	no	—	—	nan	0.577
20260819T143212353645	23.0	yes	0.111	0.117	nan	0.544
20260819T143253960183	58.9	yes	0.110	0.119	nan	0.382
20260819T143403127827	58.9	yes	0.132	0.112	nan	0.452

/home/yambox/bspline-policy/.venv/bin/python real_env/yam_teleop/rollout_local_policy.py --env yam-vrkit --policy bspline --ckpt-path /home/yambox/bspline-policy/ckpt_pick200_ee/deploy_ema.ckpt --diffusion-policy-dir /home/yambox/bspline-policy/diffusion_policy --device cpu --control-freq 100 --data-freq 10 --origin-time-scale 10 --num-inference-steps 16 --predict-before-end 0.7 --speed-up-times 1.5 --max-steps 6000 --num-episodes 5 --save --output-dir data/pick200-ee-vrkit_1.5x

data/pick200-ee-vrkit_2x

4/5 grasps.

4/5 — ckpt_path=deploy_ema.ckpt policy=bspline control_freq=100.0 data_freq=10.0 origin_time_scale=10.0 num_inference_steps=16 predict_before_end=0.7 speed_up_times=2.0 max_steps=6000 device=cpu

episode	dur (s)	grasp	z@close	lift	stall %	speed (rad/s)
20260819T143543050023	58.9	no	—	—	nan	0.213
20260819T143652023943	26.7	yes	0.119	0.120	nan	0.417
20260819T143802580789	51.2	yes	0.115	0.151	nan	0.406
20260819T143914019393	28.2	yes	0.114	0.086	nan	0.353
20260819T144006254729	36.0	yes	0.112	0.123	nan	0.414

/home/yambox/bspline-policy/.venv/bin/python real_env/yam_teleop/rollout_local_policy.py --env yam-vrkit --policy bspline --ckpt-path /home/yambox/bspline-policy/ckpt_pick200_ee/deploy_ema.ckpt --diffusion-policy-dir /home/yambox/bspline-policy/diffusion_policy --device cpu --control-freq 100 --data-freq 10 --origin-time-scale 10 --num-inference-steps 16 --predict-before-end 0.7 --speed-up-times 2 --max-steps 6000 --num-episodes 5 --save --output-dir data/pick200-ee-vrkit_2x

Videos are re-encoded to 320x240 and data.pkl is gzipped (gzip.open(...) then pickle.load). The policy consumed 84x84 crops, so nothing it acted on is lost. Re-run with --raw for originals.