albertwilcox/mesa-pi05

MESA π0.5

π0.5 (pretrained by Physical Intelligence) finetuned on MESA-70 (70 tasks; LeRobot format) for the single-arm Franka Panda setting of MESA. This is the checkpoint behind the π0.5 MESA results in the MESA paper.

This is an openpi (JAX/orbax) checkpoint; use it with our fork, pairlab/openpi-mesa, config pi05_mesa.

Model details

Base model	π0.5 (gs://openpi-assets/checkpoints/pi05_base)
openpi-mesa config	pi05_mesa
Training data	albertwilcox/mesa-70-lerobot
Training	50k steps, batch size 128, default openpi optimizer; final checkpoint (step 49999)
Cameras	left-shoulder camera (leftshoulder) and wrist camera (robot0_eye_in_hand), 224x224 RGB
Proprioception	8-D: 7 joint positions + gripper width (robot0_joint_pos, robot0_gripper_jaw_width)
Actions	8-D absolute joint-position targets (7 joints + gripper). The model predicts joint positions relative to the current state (gripper absolute); the policy's output transforms convert them back to absolute targets.
Action chunk	20 steps at 20 Hz; the MESA evaluation server executes the first 5 actions before replanning
Language	task instruction
Normalization stats	assets/mesa/norm_stats.json

Usage

Install openpi-mesa following its README, then download and serve the checkpoint:

uv run huggingface-cli download albertwilcox/mesa-pi05 --local-dir checkpoints/mesa-pi05
uv run scripts/serve_policy.py --port 8001 policy:checkpoint \
  --policy.config=pi05_mesa \
  --policy.dir=checkpoints/mesa-pi05

From the MESA repository, run the evaluation server against the same port:

uv run scripts/eval_server_parallel.py \
  --port 8001 \
  --eval-set-name mesa-70 \
  --num-rollouts-per-task 50 \
  --controller-type joint_pos

Evaluation suites: mesa-70, mesa-spatial, mesa-instance, mesa-composite, mesa-category. See the MESA documentation for details.

Results

Success rates (%) from the MESA paper (Table 2), 50 rollouts per task:

Suite	Success rate
MESA-ID	62.8
MESA-Spatial	68.6
MESA-Instance	55.4
MESA-Composite	46.7
MESA-Category	26.4
Average	52.0

Notes

• The checkpoint contains the model parameters (params/), the model config (config/), and normalization stats (assets/); optimizer state is not included.
• π0.5 was finetuned without knowledge insulation.

License

Released under the Apache 2.0 license. The model is finetuned from openpi weights that build on PaliGemma; use of the weights may also be subject to the Gemma Terms of Use.

Citation

@inproceedings{
wilcox2026mesa,
title={{MESA}: An Evaluation Framework for Compositional, Semantic, and Spatial Generalization in Robotics},
author={Wilcox, Albert and Chang, Frank and Nguyen, Nhi and Chakraborty, Aishani and Collins, Jeremy A. and Saxena, Vaibhav and Joffe, Benjamin and Karamcheti, Siddharth and Garg, Animesh},
booktitle={10th Annual Conference on Robot Learning},
year={2026},
url={https://openreview.net/forum?id=Br2rXixvyN}
}