Ballot is an XR data platform that captures full-state, physics-grounded telemetry from live virtual and mixed-reality sessions, then re-simulates those episodes for flexible observation and classifies the data for AI training. The company targets robotics and world-model teams needing rights-cleared, frame-level queryable interaction data, as well as XR studios that supply capture through its Cast SDK. Its pipeline preserves joint vectors, contact points, velocities, and game state, enabling deterministic replay from any angle or lens.
Funding
Funding not disclosed
Founders
Product
Problem
Robotics and world-model teams need large volumes of real-world interaction data to train embodied AI systems, but conventional video footage lacks the underlying physics state, joint-level telemetry, and multi-agent context required for effective learning. Existing capture methods are also difficult to rights-clear at scale, and data provenance is often unclear, limiting their usability in commercial training pipelines.
Solution
Ballot provides a full-state XR data capture and classification platform that records not just visuals but the complete physical state of an action — including joint vectors, contact points, velocities, and physics state — at a fixed tick rate. Captured episodes can be re-simulated deterministically, allowing models to observe the same moment from any angle, at any resolution, and through any filter. The Cast SDK enables studios to integrate capture directly into their games with no runtime cost, while classification happens on Ballot's side, turning raw episodes into labeled, training-ready datasets. All data is rights-cleared and audit-ready, with provenance attached at the source and frame-level queryability across video, telemetry, and state.
Target Audience
Primary buyers are robotics and world-model teams licensing multi-agent, physics-grounded, full-body VR/XR telemetry, while XR/VR studios and developers who integrate the Cast SDK serve as suppliers of rights-cleared gameplay capture.
Features
- Full-body, multi-agent, 6DoF interaction data with fixed-tick telemetry, including joint vectors, action states, contact points, velocities, and physics state
- Deterministic replay simulation that re-runs any captured episode from arbitrary camera angles, resolutions, and filters
- Cast SDK designed around the game's frame budget, enabling developer-side capture with no runtime performance cost
- Ballot-side classification pipeline that ingests raw episodes and outputs labeled sets ready for model training
- Rights-cleared and audit-ready data with provenance attached at capture, queryable at the frame level across video, telemetry, and state