Hong Kong
A look at what I’m currently building and exploring.
Exploring reinforcement learning for autonomous aerial filming, with the goal of helping a drone follow a moving subject and find suitable viewpoints. Our current work brings together target-following behavior and human-position estimation to support more responsive filming in complex environments.
Exploring how embodied agents can understand everyday instructions, interpret their surroundings, and decide where to go next. AstraNav connects visual perception with language-guided decision-making, with a focus on multi-stage tasks that require recognizing landmarks and returning to previously visited places.
Our end-to-end UAV navigation project has reached a flight speed of 120 km/h. Watch the latest flight demonstration below and the earlier demo on Bilibili. More details coming soon.