Reservoir Computers for High-Speed Predictive Error Correction of IMU Data

ResCon Technologies specializes in a low-data, low-power, high-speed Machine Learning (ML) approach known as Reservoir Computing. A Reservoir Computer (RC) can train for a complex task extremely quickly, and is so lightweight it can easily be deployed to embedded microcontrollers like those found on Inertial Measurement Unit (IMU) boards. In addition, RCs act as ideal data fusion and state estimation algorithms, though unlike Kalman Filters, their underlying parameters learn and adjust to truth data automatically. This means that the host vehicle's state estimator can change over time as the vehicle changes, optimizing the accuracy of the estimation and minimizing error. The team will build on work performed in the Phase I to test the RC approach using real-world data and actual hardware. With ResCon's lead, the Phase II project team will expand to a world-class group of performers, including Lockheed Martin Missiles and Fire Control (LMMFC), Inertial Labs, and Infinity Labs. ResCon will procure IMU hardware from Honeywell and Analog Devices, Inc (ADI) to verify results on the exact IMU simulated in the Phase I and definitively show functionality on a wide variety of devices and systems of interest to the Department of Defense (DoD). Approved for Public Release | 23-MDA-11635 (9 Nov 23)