pi05_combined_initpose_full
A pi0.5 checkpoint fine-tuned on the local/libero_combined_bowl5_initpose dataset
(see siruku6/libero_combined_bowl5_initpose),
starting from all parameters of
lerobot/pi05_libero_base,
revision a217bfd3b14673cf2ce597e69997ab21866438dd.
This repository contains two checkpoints from the same training run: 003000 and
005500. Two further checkpoints continuing training from 005500 are published
separately: siruku6/pi05_combined_ae7k
(action-expert-only, +7,000 steps) and
siruku6/pi05_combined_cont4k
(full-parameter, +4,000 steps).
Training
| Steps | 5,500 (checkpoints saved at 3,000 and 5,500) |
| Batch size | 64 |
| Trainable parameters | all 4.14B (vision encoder and VLM unfrozen) |
| Optimizer | AdamW, lr 2.5e-5, weight decay 0.01, betas (0.9, 0.95), grad clip norm 1.0 |
| Image augmentation | disabled |
| Action chunk | chunk_size=50, n_action_steps=10 |
| Dataset | local/libero_combined_bowl5_initpose (111 tasks, 21,642 episodes) |
The full configuration as actually used is in each checkpoint's train_config.json.
Evaluation
The 005500 checkpoint was compared against a checkpoint trained the same way
(all parameters, 5,500 steps, same seed) on libero_plus_bowl5 instead of
libero_combined_bowl5_initpose
(siruku6/pi05_bowl5_full) on a
Track2 validation protocol (84 rows x 5 episodes, EGL rendering):
| Training data | Total score | Success rate | Collision rate |
|---|---|---|---|
libero_plus_bowl5 (pi05_bowl5_full) |
0.4055 | 0.798 | 0.138 |
libero_combined_bowl5_initpose (this checkpoint) |
0.3515 | 0.705 | 0.171 |
Paired comparison on 420 matched trials: pi05_bowl5_full won 70, this checkpoint won
31 (p<0.001) — a clear, statistically significant gap. Two confounded factors changed at
once relative to libero_plus_bowl5 (1,532 episodes replaced by re-shot ones, and the
40 evaluated tasks diluted with 16% of frames from 71 tasks outside the evaluation set),
so this result cannot separate which one drove the gap. This dataset/checkpoint line was
not selected for further use; it and its continuations
(pi05_combined_ae7k, pi05_combined_cont4k) are published for transparency and
reproducibility of that comparison.
Modifications to the base model
model.safetensors in each checkpoint is modified relative to the base checkpoint:
it is the result of gradient updates performed by the author. No other file carried over
from the base model is modified.
License and redistribution
This is a Model Derivative of Gemma (via PaliGemma inside pi0.5) and is therefore distributed under, and subject to, the Gemma Terms of Use.
Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms
By using or further distributing these weights (or any derivative of them) you agree to:
- the Gemma Terms of Use, a copy of which governs this repository and must be passed on to any recipient you redistribute to; and
- the Gemma Prohibited Use Policy, whose use restrictions apply to this derivative and to every derivative made from it.
Reference implementation of pi0.5: openpi (Apache License 2.0). Training data is derived from LIBERO (Lifelong-Robot-Learning/LIBERO, MIT) and LIBERO-plus (Sylvest/LIBERO-plus, MIT).
Provided as-is, without warranty of any kind.