Closing the gap between simulation and the world.
Open labs advancing embodied learning, perception, high-fidelity twins, and edge intelligence — with students as co-authors, not spectators.
The Digital Twin & Deterministic Physical AI.
You can't trust an embodied system you can't reproduce. Our digital-twin program runs the same deterministic loop as the real machine — fed by live IoT telemetry, rendered in the browser with WebGPU — so autonomy can be verified before and during deployment.
The senses
Real devices stream pose, battery, LiDAR, and sensors over the wire; the twin ingests live state.
Embedded WebGPU
Rust/WASM + WebGPU render the twin at 60 fps in the browser — no app, no cloud GPU, on the edge device itself.
A live replica
A synchronized virtual copy of the asset and its world — runs forward to predict, backward to replay.
Reproducible loop
A seeded, fixed-timestep loop makes runs bit-for-bit reproducible — and flags the instant reality diverges from prediction.
A robot's eyes, on-device.
Live monocular depth estimation — the model that lets a machine judge the distance to every pixel — running entirely in your browser via WebGPU. No cloud, no install: the same embedded-edge inference a robot runs on itself. Point it at a sample scene or your own webcam and watch the depth field render in real time.
Where we push.
Four threads, one objective: autonomy that works outside the lab.
Policies that transfer
Reinforcement and imitation learning that survives the sim-to-real gap on real airframes.
Multimodal sensing
Fusing LiDAR, thermal, and RGB for robust autonomy in degraded, real-world conditions.
High-fidelity simulation
Differentiable physics and sensor models that make a twin predictive, not just pretty.
On-device intelligence
Compressing and scheduling models for real-time inference under power and thermal limits.