Added
Updated
Datoric is a San Francisco robotics training data company supplying licensed, consented datasets for physical AI. Its catalog covers egocentric task video, human manipulation recordings with synchronized depth, motion capture, and IMU streams, tactile-glove signals, UMI dual-camera bimanual demonstrations, and robot teleoperation programs on buyer-specified robots. A second tier spans agentic computer-use traces, conversational and ASR voice data, and action-conditioned gameplay for world models.
Buyers can license from the stock catalog or commission build-to-spec collection programs scoped to a task, embodiment, and sensor set. Rows carry consent, provenance, and usage terms recorded at the source, and custom programs are gated on agreed acceptance criteria before scale-up.
Frontier robotics and voice teams are hitting a data bottleneck as models move into physical environments: the volume of egocentric and manipulation data required for robotics pretraining remains orders of magnitude below what language models consumed, and no single collection method has emerged as the standard. Practices blend general video, egocentric capture, teleoperation, and simulation, with new embodiments and collection methods continuing to evolve.
Demand is shifting toward harder, more diverse tasks as common task types commoditize, which favors suppliers who can source rare scenarios under consented, verifiable conditions. Vendors that can guarantee provenance and quality at collection time, rather than filtering after the fact, are positioned to capture the premium segment of this market.
Datoric's core advantage is provenance-first sourcing: consent, licensing scope, and data origin are defined as part of each collection specification and made reviewable with the delivered dataset, rather than reconstructed after the fact. Its collection runs through private, project-specific contributor systems separated by modality, customer, and trust boundary, with no public marketplace or open upload surface.
The company reports more than 300,000 active contributors recording through its private apps, with every submission linked to its creator and consent record, and applies spec-driven QA covering signal and metadata checks, redaction review, labeling consistency, and acceptance against agreed criteria. Buyers evaluating robotics data can review dataset records, capture specifications, and published limitations through the public catalog rather than relying on vendor claims.
Datoric does not publish list prices. Its stock catalog is licensed for commercial AI training under engagement-specific direct agreements, with the signed agreement controlling scope, and licensing documentation included in the review package buyers receive before delivery.
Custom collection programs follow the same motion: a pilot against agreed acceptance criteria must pass before the program scales, implying sales-led enterprise pricing rather than self-serve rates. No public pricing page exists on any official surface walked.
Datoric collects licensed, consented training data for physical AI and robotics.
