Cornell's REACH keeps underwater robots swimming safely

Cornell University researchers have unveiled REACH (Real-time Estimator of Actuator Control and Health), a fault prediction system designed for soft underwater robots modeled after eels. Published on arXiv, this algorithm addresses a critical challenge in soft robotics: actuator degradation in harsh marine environments. The system combines a soft robot physics model with a sigma point filter and statistical hypothesis testing to detect actuator failures in real time.

In testing, REACH successfully predicted actuator health across three swimming gaits—linear swimming, wide turning, and tight turning—using IMU or bend sensor data. The algorithm proved robust even with noisy sensor data and variations in manufacturing tolerances. Sensor placement analysis showed that two IMUs are sufficient for fault detection, while bend sensors require three sensors for comparable performance. The researchers validated the system's statistical consistency in identifying actuator degradation, demonstrating its potential for autonomous underwater missions where real-time fault detection is critical.