π0.5 — BimanualYAM Cups

Task: "Stack the cups." — 100 episodes, 39,184 frames @ 30 fps (RoboColosseum/BimanualYAM-datasets/Cups)
Inputs/outputs: cameras top / left wrist / right wrist + instruction; 14-D absolute joint state and action
Base: gs://openpi-assets/checkpoints/pi05_base (openpi, JAX)
Trained: Full fine-tune, EMA 0.99 (config pi05_yam, recipe of pi05_aloha_pen_uncap, adapt_to_pi=False)
Optim: AdamW, cosine, peak LR 2.5e-5 → 2.5e-6, 1k warmup; GBS 64 (16/GPU x 4); 5 epochs; fp32 master weights, bf16 compute; 4x H100; wall time 1h39m
Files: Final checkpoint in openpi format: params/ (EMA weights) and assets/<dataset>/norm_stats.json. Optimizer state not included.
Load: openpi.policies.policy_config.create_trained_policy(get_config("pi05_yam"), <this folder>) (config: configs/openpi/yam.py)
Code: RoboColosseum harness (scripts/train/pi05.sh); logs in W&B project RoboColosseum

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including RoboColosseum/yam-cups-pi0.5