vikram-avea/yam-physical-ai-hack-640x480

YAM Physical AI Hack Dataset (640x480@30fps)

LeRobot v3.0 dataset for the Physical AI Hack event featuring YAM robots (LimX Sentinel humanoids).

Dataset Info

• —Robot: YAM (LimX Sentinel humanoid)
• —Resolution: 640x480
• —FPS: 30 (TRUE 30fps, re-encoded for consistency)
• —Total Episodes: 47 (32 + 15)
• —Total Frames: 43,252
• —Task: Teleoperation demonstrations
• —Format: LeRobot v3.0

Why This Dataset?

This dataset was specifically prepared for the Physical AI Hack 2026 event with: ✓ True 30fps - Re-encoded from mixed 30/31fps sources for consistent timestamps ✓ Optimal resolution - 640x480 balances quality and training speed ✓ Proper alignment - Accurate action-observation temporal relationships

Recording Sessions

Merged from two recording sessions:

• —Session 1: 32 episodes (re-encoded from 31fps to 30fps)
• —Session 2: 15 episodes (native 30fps)

Usage

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

# Load dataset
dataset = LeRobotDataset("vikram-avea/yam-physical-ai-hack-640x480")
print(f"Episodes: {len(dataset)}")
print(f"Frames: {dataset.num_frames}")  # 43,252
print(f"FPS: {dataset.fps}")  # 30

# Training example
from torch.utils.data import DataLoader
dataloader = DataLoader(dataset, batch_size=32, shuffle=True)
for batch in dataloader:
    obs_state = batch["observation.state"]  # [32, 12]
    obs_wrist = batch["observation.images.robot1_wrist"]  # [32, 3, 480, 640]
    obs_overhead = batch["observation.images.overhead"]  # [32, 3, 480, 640]
    action = batch["action"]  # [32, 8]
    # Your training code here

Features

• —observation.state: Float32[12] - Joint positions (6 arm + 6 gripper states)
• —action: Float32[8] - Joint commands (6 arm + 2 gripper commands)
• —observation.images.robot1_wrist: Uint8[3, 480, 640] - Wrist RGB camera
• —observation.images.overhead: Uint8[3, 480, 640] - Overhead RGB camera

About YAM Robots

YAM robots are LimX Sentinel humanoid platforms for dexterous bimanual manipulation. Collected during Physical AI Hack 2026.

Citation

@dataset{yam_physical_ai_hack_2026,
  title={YAM Physical AI Hack Dataset},
  author={Avea Robotics},
  year={2026},
  publisher={HuggingFace},
  howpublished={\url{https://huggingface.co/datasets/vikram-avea/yam-physical-ai-hack-640x480}}
}