Datoric
Datoric provides the foundational data layer for physical AI by identifying the data-driven causes behind model failures and capability gaps. The company runs controlled collection experiments across video, motion, and tactile modalities, using treatment-effect estimation to validate data interventions before production. Its available egocentric residential video dataset spans 80 non-cooking task categories with structured annotations and privacy-redaction review.
- Artificial Intelligence
- Robotics
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
AI models operating in the physical world often fail unpredictably, and the root causes of these failures are frequently data-driven rather than algorithmic. Teams struggle to identify which data characteristics—such as missing edge cases, annotation gaps, or sensor limitations—contribute to model errors, leading to costly trial-and-error iteration. Without a systematic method to trace failures back to data and validate fixes, development cycles slow and deployed systems remain unreliable.
Solution
Datoric provides the foundational data layer for AI in the physical world by identifying the data-driven causes behind model failures and capability gaps. The company runs controlled collection experiments across video, motion, and tactile modalities, systematically varying data characteristics to estimate their effect on downstream performance. Internal systems apply treatment-effect estimation to quantify how specific data changes impact model behavior, then downstream evaluation tests whether an intervention transfers before it is specified for production. This empirical approach turns model failures into validated data methods, enabling teams to procure or collect data with confidence that it will resolve their specific issues. Datoric also offers pre-built datasets, such as egocentric residential video with 80 non-cooking task categories, complete with structured annotations and provenance documentation.
Target Audience
Primary customers are physical AI companies and research teams developing robots, autonomous systems, or multimodal perception models that require validated, purpose-built training data to address specific failure cases and capability targets.
Features
- Controlled collection experiments across video, motion, and tactile modalities with treatment-effect estimation to isolate data-driven causes of model failure
- Downstream model evaluation pipeline that validates whether a data intervention transfers before production specification
- Egocentric residential video dataset covering 80 non-cooking task categories (cleaning, laundry, organizing, dish handling, pet care, plant care, home maintenance) with 1080p+ first-person capture
- Rich annotation schema including task phases, hand-object interaction events, room/object/tool/surface labels, before/after state changes, and hand visibility metrics
- Provenance and rights chain with contributor consent documentation, commercial licensing, and privacy-redaction review (PII redaction, face blurring, identifier removal)
- Flexible delivery options: buyer review package with real video and JSON annotations, plus custom scoping by task category, environment, annotation needs, and hours