AirVLA FT-C (pi0, step 15000)

The strongest pure VLA policy of the campaign, and the policy used by both terminal-control hybrids.

Part of the AirVLA MSc dissertation campaign: adapting a pretrained pi0 vision-language-action policy to a quadcopter with a 2-DoF arm and parallel gripper, evaluated in MuJoCo 3.3.4.

Property Value
Base model lerobot/pi0
Training data hanapasta/airvla_v21 (v2 + F1 terminal-corrective, F1 up-weighted)
Initialised from lerobot/pi0 base
Training steps 15,000
Checkpoint selection lowest pinned-noise validation MSE over the saved ladder

Frozen-protocol result (n = 60 pick + 20 navigation, paired scenes)

1/60 grasped - 47/60 correct-target approaches - 145.8 mm median miss (87.6 mm target-true) - 12/20 navigation.

Provenance note: this is a deterministic retrain of the original FT-C checkpoint, which was lost to a storage-cleanup error. It was accepted under equivalence gates: pinned-noise validation MSE 7.7e-5 vs the original 7.679e-5 (0.3%), plus a matching closed-loop evaluation. See the ledger for the full incident record.

Usage

Evaluated with the frozen harness eval_v2.py from the code repository:

python eval_v2.py <this_checkpoint> 60 20 --torchseed 1000 --tag run --video

Code, reproduction guide and the full experimental ledger: https://github.com/robotics-hana/drone-version2

Evaluation uses MuJoCo 3.3.4; a different simulator version changes contact behaviour enough to invalidate comparison with the banked results.

Downloads last month
13
Safetensors
Model size
4B params
Tensor type
F32
·
BF16
·
Video Preview
loading