---
license: apache-2.0
tags:
  - robotics
  - vla
  - pi0.5
  - expo-ft
  - maniskill
  - lora
---

# ExpoFT π₀.₅ — PushCube-v1 (SFT, checkpoint 600)

LoRA-finetuned checkpoint of [π₀.₅](https://www.physicalintelligence.company/blog/pi05) (`pi05_droid_jointpos`),
adapted to ManiSkill's `PushCube-v1` task as part of the [ExpoFT](https://pd-perry.github.io/expo-ft/)
reproduction project.

- **Success rate**: 62% on 200 held-out fixed-seed evaluation episodes
  (SE = sqrt(p(1-p)/n))
- **Training**: LoRA supervised fine-tuning on 50 motion-planning demonstrations
- **Action space**: joint-space (`pd_joint_delta_pos`), matching pi0.5's DROID
  pretraining distribution
- **Normalization stats**: official DROID stats (`gs://openpi-assets/checkpoints/pi05_droid_jointpos/assets/droid`)

## Warning: JAX only

This checkpoint is in its native **JAX/orbax** format (as produced by
[`openpi`](https://github.com/Physical-Intelligence/openpi) training) and has
**not** been converted to PyTorch. `openpi`'s official conversion script
(`examples/convert_jax_model_to_pytorch.py`) exists, but PyTorch-side LoRA
support is currently absent in `openpi`, and unresolved upstream issues
report output mismatches between JAX and converted PyTorch weights for
pi0.5 specifically -- so no PyTorch conversion is provided here.

## Usage

Point `pi05_weight_loader_path` at the downloaded `params/` directory when
building the model via `openpi`'s `CheckpointWeightLoader`, using the
`expo_pi05_droid_lora_finetune_sft_joint_state` training config as the base.

## Code

Training/eval pipeline: [Josh012006/expo-ft](https://github.com/Josh012006/expo-ft)