SUTD researchers have developed a reinforcement-learning-based safety system that teaches a stair-traversing service robot to brace itself mid-fall, addressing one of the biggest barriers to deploying autonomous robots on staircases.

A new study in Artificial Intelligence and Autonomous Systems shows that neural networks pre-trained to predict continuous gait cycle progression can be reused to improve gait phase classification from foot-mounted inertial sensor data. The approach achieved an F1-score of 0.9788, CPU inference latency below 0.07 ms with a compact DNN, and 92.3% accuracy on an independent young-adult dataset.