HOT3D: Multi-View 3D Hand & Object Tracking for Openclaw

A state-of-the-art multi-view 3D tracking system from Meta Facebook Research optimized for egocentric hand-object interactions on XR devices.

fujunhy
v0.1.0
Mar 14, 2026
0
853
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install hot3d

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install hot3d using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).

Prefer to download?

Get the raw skill files in a ZIP archive.

What is HOT3D: Multi-View 3D Hand & Object Tracking?

HOT3D is a comprehensive tracking framework designed for Meta Aria smart glasses and Quest VR headsets. As a core component of the Openclaw Skills ecosystem, it provides precise 3D world coordinates for 21-keypoint hand poses and 6DOF object tracking during complex manipulations. Unlike standard single-view solutions, HOT3D leverages multi-view egocentric data to achieve millimeter-level accuracy in 3D space.

This skill enables developers to reconstruct hand meshes and align object models within a synchronized multi-camera environment. By utilizing this Openclaw Skills entry, researchers and developers can access robust visualization tools for 3D skeleton projections and trajectory analysis, making it the definitive choice for high-fidelity XR interaction development.

HOT3D: Multi-View 3D Hand & Object Tracking Use Cases

  • Implementing high-precision hand-object interaction analysis in XR environments.
  • Research-grade benchmarking for 3D hand pose estimation algorithms.
  • Developing applications for Meta Aria smart glasses and Quest headsets requiring world-coordinate tracking.
  • Analyzing contact points and grasp types in daily activity sequences.
  • Creating rich 3D overlays and skeletal visualizations on multi-view video streams.

How HOT3D: Multi-View 3D Hand & Object Tracking Works

  1. The system ingests multi-view RGB-D video streams synchronized across 3-5 cameras.
  2. Keypoint detection identifies 21 specific 3D hand joints with confidence scoring for each view.
  3. Object pose estimation calculates the 3D position and orientation (quaternion/rotation) of manipulated items.
  4. Hand-object interaction logic detects contact points and classifies manipulation phases.
  5. The final output generates 3D coordinates in a unified world frame, suitable for rendering meshes or skeletal overlays.

HOT3D: Multi-View 3D Hand & Object Tracking Setup

To integrate this Openclaw Skills module, follow these installation steps:

# Clone the HOT3D repository
git clone https://github.com/facebookresearch/hot3d.git
cd hot3d

# Install required dependencies including PyTorch3D and Open3D
pip install -r requirements.txt

# Note: Accessing the full dataset requires registration at the official project page

HOT3D: Multi-View 3D Hand & Object Tracking Data Schema & Taxonomy

HOT3D organizes tracking data through a structured frame-based taxonomy:

Data Type Description Format
Hand Pose 21-keypoint 3D joints (Left/Right) 21x3 Array (x, y, z)
Object Pose 6DOF position and rotation matrix [x, y, z] + 3x3 Matrix
Metadata Camera intrinsics and confidence scores JSON / Dict
Visuals 3D meshes and skeletal projections OBJ / PNG Overlay

HOT3D: Multi-View 3D Hand & Object Tracking Advanced Features

  • Multi-agent support for tracking simultaneous interactions across different users.
  • Temporal consistency algorithms to maintain tracking stability across high-frequency frames (30-60 Hz).
  • Integration hooks for Aria SDK and OpenXR platforms.
  • Automated force estimation capabilities when combined with auxiliary sensor data.
  • Comprehensive evaluation suite via the hand-tracking-toolkit for metric analysis.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*