Edge inference
Small, structurally constrained dynamics models running onboard instead of on a GPU. At around 900K parameters versus 15M, a world model fits on a Jetson or an ESP32-class companion processor. The natural use on Eidon-style hardware is predictive tracking: anticipating where a limb is going a few hundred milliseconds ahead, smoothing noisy sensors, fusing IMU and visual signals into one state estimate.
Connections
Parent: Robotics
Tangents
- Physics-structured models: few parameters, fits on-device
