Physics-accurate simulation and adaptive deep reinforcement learning that predict play as it happens — live shot tracing, strategy insight, and broadcast overlays for complex sports, starting with curling.
Get In TouchExisting analytics rely on rules-based and static machine-learning models, so real-time prediction simply isn't available to broadcasters — and viewer engagement is more important than ever.
Regression and static ML limit predictive accuracy in real time. Complex sports like curling can't be forecasted without models that adapt shot to shot.
Existing solutions such as golf shot tracking are hardware-intensive, costly to install per event, and hard to extend to other sports.
Streaming has fractured audiences. Broadcasters lack scalable, AI-driven overlays and insights that deepen fan engagement across multiple sports.
FrostPath fuses a full-fidelity physics engine, computer-vision intake, adaptive DRL agents, and broadcast-ready 3D graphics — a complete path from live camera feed to on-air insight.
A detailed 40 m curling sheet model computing 24,000 stone-and-pebble interactions per second, built on the most widely accepted friction physics research.
Overhead-camera rock position capture — developed in partnership with the Grand Slam of Curling — feeds real games straight into the simulation, no on-ice hardware required.
Two deep reinforcement learning curling agents with an adaptive framework that re-learns conditions end to end — and even shot to shot.
A 4K curling arena VFX model built in Blender, Unity, and NVIDIA Omniverse, ready for live broadcast integration.
Curling is uniquely hard to model: the playing surface itself changes throughout the game, and the value of a shot is rarely obvious. That's exactly why it demands adaptive deep reinforcement learning.
Sheet mechanics change every broadcast, game, end — even rock to rock: pebbling, temperature, ice maintenance frequency, directional sweeping, individual sheet lean and tilt, and shot-lane usage all shift conditions continuously.
Draw, hit, raise, or freeze? The optimal call requires subtle strategy, and true shot value can only be computed after the end is scored — 16 shots later. Adaptive agents learn to evaluate options where static models can't.
In development with broadcast partners — television overlays and advanced graphics driven live by the simulation:
Predicted and actual rock paths with on-screen speed as the stone travels, rendered live.
Quantify what sweeping is really doing.
Strategic choices and their expected value.
Advertisement placements integrated into broadcast graphics.
Incorporate overhead-camera match datasets into the adaptive DRL training loop.
Continue development of TV overlays and advanced graphics — in flight now.
End-to-end live streaming integration test with broadcast partners.
Advanced curling analytics for coaches and professional teams.
Broadcasters, leagues, and teams — pilot with us and lead your sport.