Sureform gathers high-quality multimodal human interaction data across varied environments to support the development of advanced world models and physical AI systems. The dataset is offered to AI research teams and enterprises building embodied agents, enabling more accurate simulation and perception capabilities. Access to the data is provided through subscription or licensing agreements, allowing customers to integrate the curated data directly into their training pipelines.
Funding
Funding not disclosed
Founders
Product
Problem
Training robust world models and physical AI systems requires large-scale, high-fidelity multimodal data that captures human interactions in real-world settings. Existing datasets are often limited to single modalities, synthetic environments, or narrow contexts, which hampers generalization and slows progress in autonomous perception and manipulation.
Solution
Sureform operates a data‑collection platform that captures synchronized video, audio, depth, inertial, and interaction metadata across diverse indoor and outdoor environments. The pipeline includes calibrated sensor rigs, automated quality control, and expert annotation to produce consistently labeled multimodal recordings. Collected data are packaged in standardized formats (e.g., ROS bags, TFRecord) and made available through a secure API and bulk download options. By licensing these curated datasets, AI research labs, robotics firms, and enterprises can train and evaluate world‑model architectures, sim‑to‑real transfer pipelines, and embodied agents without building their own data infrastructure. Sureform also offers custom data‑capture projects to fill niche scenario gaps, ensuring that customers receive domain‑specific samples aligned with their development roadmaps.
Target Audience
Primary customers are AI research laboratories, robotics manufacturers, and enterprise teams building autonomous perception or manipulation systems that require real-world multimodal training data.
Features
- End‑to‑end sensor suite (4K RGB, binaural audio, LiDAR, IMU) with time‑synchronized streams for precise multimodal alignment
- Automated data validation pipeline that flags motion blur, audio clipping, and sensor drift before storage
- Expert annotation workflow covering object bounding boxes, action labels, affordance tags, and environmental context metadata
- Open‑format delivery (ROS2 bag, TFRecord, Parquet) with versioned schema to simplify integration into training pipelines
- Secure, token‑based API for programmatic access, supporting incremental downloads and streaming of large datasets
- GDPR‑compliant data handling and de‑identification tools to protect privacy while preserving research utility
- Custom capture services for industry‑specific scenarios (e.g., warehouse logistics, human‑robot collaboration)
- Documentation and SDKs (Python, C++) for seamless ingestion into popular ML frameworks (PyTorch, TensorFlow, JAX)