Parameter	Value
Base model	pi0.5 (PaliGemma 2B + Gemma 2B action expert)
Total parameters	~3.35B
Action dimension	32
Action horizon	10
Batch size	16
Training steps	5,000
Learning rate	Cosine decay: warmup=500, peak=5e-5, end=5e-6
Optimizer	AdamW (gradient clip norm=1.0)
GPUs	8x NVIDIA A100

Checkpoints

Task: Object Arrangement
Training data: Blue-Black-Red order only
Dataset: real_arrangement_blue_black_red
Robot: Franka Panda (7-DOF)
Cameras: Base RGB + Wrist RGB (256x256)

Part of Mode Editing Research

This checkpoint is part of the "Don't Filter Your Data, Edit Your Policy" project (CoRL 2026).