A.I. at the Edge of the Edge: Intelligent Sensor Falling Detection for Humanoid Robot

This work addresses the problem of fall detection for humanoid robots using an Intelligent Sensor Processing Unit (ISPU). The ISPU offloads the computational burden of robot movement tracking from the main microcontroller. An ad hoc dataset was created and used to train a Convolutional Neural Network (CNN) model for activity recognition. To accommodate the ISPU’s limited RAM, the CNN model was quantized. The system detects falls in real-time, enabling the robot to activate specific protection strategies based on the predicted fall direction. This innovative approach brings AI to the extreme edge, leveraging intelligent sensors to classify robot behaviors and reduce computational load on the microcontroller.