pi0.5 Build Block Tower — Baseline
Fine-tuned pi0.5 checkpoint for building a block tower, trained with imitation learning only (no advantage conditioning) on 200 human demonstrations.
Experiment
- Objective: Train a block tower policy from pi0.5 base weights using human demos only (imitation learning baseline).
- Weight init:
weights/pi05_base/params(pi0.5 base weights). - Advantage conditioning: None (standard imitation learning).
- Target steps: 100,000
Config
- Config name:
pi05_build_block_tower_baseline - Model: pi0.5 (
pi05=True,action_horizon=50) - Batch size: 36
- Learning rate: 5e-5 cosine decay (10k warmup)
- Optimizer: AdamW (gradient clip norm 1.0)
- EMA decay: 0.999
- Delta actions: enabled
- State/action space: 7D joint-space
Dataset
villekuosmanen/build_block_tower(200 episodes, LeRobot v2.1)
Checkpoint Hashes
Verify integrity with tar cf - -C checkpoints/<step> params | sha256sum.
| Step | Loss | SHA-256 |
|---|---|---|
| 25,000 | 0.0111 | e137fa5a690e21f275e9dcbbffda6428c0eabd0100270c407ac366c225674ea2 |
| 50,000 | 0.0077 | 51e01160ab7119dbbf6b7c4515900cc794eb11773c47ced6f853e0c4f976c652 |
| 55,000 | 0.0068 | 158ff29591982484b6594afaabec078bdf8b59e7b72a4c886d4a542233581d11 |
W&B
Repo Structure
assets/ # Norm stats for inference
checkpoints/<step>/params/ # Model weights (params only)
README.md # This file
TRAINING_LOG.md # Training log
Usage
from openpi.training.config import get_config
from openpi.serving.policy_server import PolicyServer
config = get_config("pi05_build_block_tower_baseline")
server = PolicyServer(config, checkpoint_path="checkpoints/<step>/params")