Embodied Intelligence Data Science
The Egocentric
Data Engine
Scaled real-world egocentric demonstrations for high-quality VLA training data infrastructure
Explore Our SolutionsData Annotation
Egocentric Operation Data
Capturing different people performing different tasks across varied real-world environments — reflecting the natural diversity of real human behavior.
Home Scene
Capturing everyday household cleaning and organization behaviors
Kitchen Scene
Record common kitchen and cooking-related actions
Wall Surface
Capturing wall painting and installation-related operations
Market Scene
Capturing market transactions and product selection actions
Cleaning Scene
Recording mopping actions in household and hotel environments
The annotation design directly supports semantic-action alignment for VLA and world model training.
Industry Gap
- ●Limited access to continuous, natural manipulation data.
- ●Existing datasets are staged or task-constrained.
- ●Simulation and teleoperation lack real-world diversity.
POV Advantage
- ●Egocentric POV captures authentic human interactions.
- ●Long-horizon trajectories reflect real task execution.
- ●Collected from everyday workflows at scale.
FirstMove pioneers a “body-agnostic” data collection paradigm, transforming every touch, every action, and every interaction humans perform in real-world scenarios into high-quality datasets that robots can learn from.
Cloud AI Platforms
Possess powerful cloud-based model capabilities but critically lack real-world physical interaction data to train embodied intelligence systems.
Robot OEMs
Require massive volumes of scenario-specific data to improve dexterous manipulation success rates and mobile platform task performance.
Research Institutions
The maturation of multimodal foundation models makes it possible to translate human action data into robot-executable instructions at scale.
02 — Technology
Training-Ready 3D Human Behavior
From raw egocentric video to temporally consistent 3D pose and action representations
Egocentric Capture
Head-mounted cameras record continuous first-person manipulation sequences in real environments.
3D Pose Extraction
Multi-view reconstruction produces temporally consistent 3D hand and body pose trajectories.
Semantic Annotation
Action labels, object states, and task descriptions are aligned to each trajectory segment.
VLA-Ready Output
Structured datasets formatted for direct ingestion by Vision-Language-Action model pipelines.
03 — Team
World-Class Interdisciplinary Team
Bringing together top talent in robotics, AI algorithms, foundation models, organizational management, and data assets to build a core leadership team with global vision and deep industrial expertise.
Daoling Song
Co-founder and COO of Zhejiang Humanoid Robot Innovation Center. Forbes China 30 Under 30. Deep expertise in the robotics industry with proven track record in humanoid robot commercialization and strong industry influence.
Chang Liu
Assistant Professor at Peking University School of Engineering. Ph.D. from UC Berkeley. Former core engineer at NVIDIA Autonomous Driving. World-class academic credentials and industrial experience in embodied intelligence and robot control.
Jack
Former P10-level foundation model lead at a major tech company. Among China's first generation of large model training experts. Led pre-training and fine-tuning of hundred-billion-parameter models with extensive industry connections.
Frank
Former senior executive at a leading national asset management firm. Over ten million square meters of multi-format property development and asset management experience. Leads the Thousand-Household Program and global crowdsourcing pipeline.
Clark
Renowned tech investor and author. Expert in AI data asset monetization and global strategy. Oversees company strategy, brand, and international operations. Published works on AI and organizational transformation.
04 — Vision
The Future of Embodied Intelligence
We believe that true embodied intelligence will not emerge from simulators alone, but from every touch, every action, and every interaction in the real world.
FirstMove’s mission is to become the most critical data infrastructure of the embodied intelligence era — continuously acquiring high-quality scarce data, continuously optimizing model capabilities, enabling every robot to understand and navigate the real world.
This is not merely a company’s vision — it is the starting point of an era.
05 — Contact
Get in Touch
Whether you are a cloud AI platform seeking high-quality embodied data, or a robot OEM requiring real-world scenario data, we look forward to accelerating the path to embodied intelligence together.