Datasets:

AnodeAI
/

Humanoid-Kinetic-Fleet

task_categories:
  - text-classification
license: apache-2.0
tags:
  - code
  - Humanoid
  - 6D pose
  - point cloud
  - Robotics
size_categories:
  - 1k-10k

🤖 Anode AI: Humanoid Kinetic Fleet (v1.0)

High-Fidelity Synthetic Tensors for Next-Gen Humanoid Perception & Control.

Anode AI’s Humanoid Kinetic Fleet is a mathematically deterministic synthetic dataset designed to bridge the Sim2Real gap for domestic and industrial humanoid robotics. Unlike standard computer vision datasets, this collection includes full 6-DoF ground truth, kinematic torque vectors, and Gaussian stochastic noise modeled on real-world 24GHz radar and LiDAR interference.

📊 Dataset Summary

Total Records: 1,240,000+ Frames
Format: .jsonl.gz (Compressed JSON Lines)
Capture Rate: 90Hz (Temporal Coherence)
Domain: Domestic Environments (Kitchen, Living Room, Dining)
Physics Engine: Anode Mud Engine v2.1 (Euler Integration)

🛠 Data Structure & Schema

Each record contains a multi-modal snapshot of the robot's state and its environment.

1. Robot Kinematics

6-DoF Pose: Precise [x, y, z] and Quaternions for the base and end-effectors.
Joint Dynamics: 18-axis joint angles and velocities.
Force Feedback: Torque vectors (Nm) and gripper pressure (N).

2. Semantic Intelligence

Object Metadata: Includes mass_kg and kinetic_energy_j for interaction logic.
Intent Prediction: Behavioral labels for dynamic entities (e.g., Child_5yo_Running).
Threat Vectors: Closing speeds and potential impact time calculations.

3. Sensor Fidelity (Stochastic Layer)

Gaussian Noise: Modeled via Box-Muller transforms to simulate sensor jitter.
Domain Randomization: Variable lighting (Lux), texture shifts, and color variations.

🔬 Technical Specifications

Parameter	Specification	Logic
Noise Model	Gaussian (Box-Muller)	Sustainable Real-World Noise
Physics Integration	Euler (dt=0.1s)	Kinematic Continuity
Integrity Check	SHA-256	Cryptographic Data Provenance
Coordinate System	RHS (Right-Handed)	Standard Robotics Convention

🚀 Usage

This dataset is optimized for:

Reinforcement Learning (RL): Training humanoids for object manipulation using mass/torque metadata.
Edge-Case Detection: Testing model failure points in low-light/high-clutter scenarios.
Sensor Fusion: Aligning 24GHz Radar returns with LiDAR point clouds.