Minghui (Scott) Zhao, a third-year PhD student in our lab, won second place in the Student Research Competition at the 30th ACM International Conference on Mobile Computing and Networking (MobiCom’24) held in Washington, DC. The award recognizes his team’s work on connecting foundation models with the physical world through reconfigurable drone systems.
The winning project, EmbodiedRDA (Embodied Reconfigurable Drone Agent), tackles a fundamental challenge in AI: enabling foundation models to interact with the physical environment. While current AI systems excel at processing text and digital content, they often struggle with real-world tasks due to limited sensor configurations and static deployment locations.
EmbodiedRDA introduces a novel solution through a custom-designed drone platform that can autonomously swap different sensors and actuators based on task requirements. The system features:
- A modular drone design that allows autonomous reconfiguration of sensors and actuators
- A ground station built from a modified 3D printer for precise payload management
- An foundation-model-powered agent that processes user commands and controls the drone’s actions
The system demonstrated practical applications such as finding optimal room temperatures, detecting chemical spills, locating objects, and delivering items. By combining foundation models with reconfigurable hardware, EmbodiedRDA represents a step toward more adaptable and capable AI systems that can actively engage with the physical world.
This recognition follows the project’s success in the conference’s demo track, where it received the Best Demo Runner-Up Award.