Downloads · 30 days
0
pi05_libero10_ipec_seed43_step2k
jiankimr/pi05_libero10_ipec_seed43_step2k
pi05_libero10_ipec_seed43_step2k is a robotics model from jiankimr. Use it for the robotics task on the model card, and read the license before you ship it in a product. It is set up for openpi. The card lists the license as apache-2.0.
Clean (un-poisoned) pi0.5 (openpi) fine-tuned on LIBERO-10 only (IPEC-COMMUNITY/libero10nonoops1.0.0lerobot, 379 episodes / 101,469 frames @ 20 Hz). Victim policy for fatigue-injection / defense-evasion experiments (L…
Access
Public
Updated Sep 14, 2026
Repo size
12.4 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
Other12.4 GB · 100%
At a glance
- Task
- Robotics
- Library
- openpi
- License
- apache-2.0
- Access
- Public
- Created
- Sep 14, 2026
- Updated
- Sep 14, 2026
- SHA
- 01b6e509
Model card
From the Hugging Face model README
pi0.5 fine-tuned on LIBERO-10 — seed 43, step 2,000 (≈5.0 epochs)
Clean (un-poisoned) pi0.5 (openpi) fine-tuned on
LIBERO-10 only (IPEC-COMMUNITY/libero_10_no_noops_1.0.0_lerobot, 379 episodes / 101,469 frames @ 20 Hz).
Victim policy for fatigue-injection / defense-evasion experiments (LPF, jerk / COBALT jitter).
| Training seed | 43 |
| Step | 2,000 |
| Epochs | ≈5.0 (= step × batch 256 / 101,469 frames) |
| LIBERO-10 SR | not evaluated (10 tasks × 10 trials = 100 episodes, 500 Hz physics, OSC_POSE) |
| Base model | gs://openpi-assets/checkpoints/pi05_base |
| Recipe | openpi pi05_libero defaults: batch 256, FSDP over 2×H100, AdamW, cosine LR (warmup 10k, peak 5e-5, constant after), EMA 0.999 |
| Action space | 7-D EE delta (pos xyz, axis-angle xyz, gripper), chunk 8 @ 20 Hz |
| Contents | params/ (orbax, EMA weights), assets/.../norm_stats.json, _CHECKPOINT_METADATA. train_state/ (optimizer) omitted. |
SR saturation — same run (seed 43), all uploaded checkpoints
| step | exact | epochs | SR (100 ep) |
|---|---|---|---|
| 2k | 2,000 | 5.0 | not evaluated |
| 4k | 4,000 | 10.1 | not evaluated |
| 6k | 6,000 | 15.1 | not evaluated |
| 8k | 8,000 | 20.2 | not evaluated |
| 10k | 10,000 | 25.2 | 89% |
Training was stopped at 10k (≈25 epochs — the same epoch budget as openpi's official 30k-step recipe on the 4-suite
LIBERO dataset, ≈273k frames); seed 42 showed SR plateaus from 10k. Only 10k is evaluated so far. Each row is its own repo: jiankimr/pi05_libero10_ipec_seed43_step<k>.
Load with openpi
from openpi.policies import policy_config
from openpi.training import config as _config
import rsec_pi05.configs # registers pi05_libero10_ipec (openpi LiberoDataConfig re-keyed for the IPEC dataset)
cfg = _config.get_config("pi05_libero10_ipec")
policy = policy_config.create_trained_policy(cfg, "CKPT") # dir containing params/ and assets/
- Task
- Robotics
- Library
- openpi
- License
- apache-2.0
- Created
- Sep 14, 2026
- Updated
- Sep 14, 2026