AIBullisharXiv – CS AI · 3h ago7/10
🧠
Beyond Motion Primitives: Behavioral Activity Recognition from Head-Mounted IMU
Researchers from Harvard's AI and Robotics Lab have developed HiT-HAR, a hierarchical deep learning model that enables AR smart glasses to recognize complex human behaviors beyond basic motion primitives using only head-mounted IMU sensors. The team created a 160K-sample dataset and demonstrated that architectural choices exploiting temporal context outperform simple model scaling, advancing the feasibility of always-on behavioral context awareness for augmented reality applications.