Downloads · 30 days
8
26% of all-time downloads
Travor278/pi05-piperx-full558-hil-bc-r1
pi05-piperx-full558-hil-bc-r1 is a robotics model from Travor278. 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 lerobot.
This is the matched data-only checkpoint from the first Evo-RL offline policy-improvement ablation for the task:
Downloads · 30 days
8
26% of all-time downloads
All-time downloads
31
Public
Parameters
3.6B
40.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors40.5 GB · 100%
From the Hugging Face model README
This is the matched data-only checkpoint from the first Evo-RL offline policy-improvement ablation for the task:
Insert the copper screw into the black sleeve.
v2sam-ego2exo-official-v3-r5-4gpujob-c52f4460-5f3b-4aa2-87bf-09cf15a33a90FORMAL_POLICY_PASS10000d85c7cd84060a924b6ef10d055491c500c5714a21ddbb49e1dbdc828b7a741476f2db449a21e1bac750b996f2e27cac6739aa63f64 (4 × 16)202609020.25acp.enable=false)1450The model started from the same task-specific full558 baseline as the paired RL-aware run, while the optimizer and scheduler were reinitialized. It uses the same mixed replay data and sampling ratio but treats executed actions as ordinary demonstrations and does not use intervention labels in the policy loss.
pretrained_model.training_state/ contains optimizer, scheduler, RNG, and step state for reproducibility.provenance/ contains the run manifest, reports, configuration, and SHA256 records.For inference, point LeRobot at the repository root or at the downloaded root directory. Do not point inference at training_state/.
The training, distributed-runtime, numerical, and checkpoint gates passed. Offline held-out comparison and real-robot Base/Data-only/RL-aware evaluation remain required before making performance claims. This checkpoint should not be interpreted as proof of improved success rate until those evaluations are complete.