CCES researcher, Xiaotong Zhang’s, PhD research at MIT has been featured on MIT News.
He assisted in the development of a novel robotic system, named “Relevance”, that can intelligently identify and prioritize objects most relevant to assisting humans in various tasks, significantly improving the efficiency, safety, and intuitiveness of robots in real-world applications.
Relevance is quantified through a novel and flexible framework that enables critical robotic functions often overlooked in previous research. This framework dynamically adjusts perception and relevance determination frequencies, allowing optimal computational resource allocation for faster perception and action generation. Its flexibility allows for the integration of diverse information sources— including human factors, scene cues, task models, and knowledge from large language models (LLMs)— to facilitate intelligent and seamless assistance. Additionally, it ensures cue sufficiency to prevent faulty actions and employs a hierarchical scene representation based on object classes and elements.
Read the full interview here.
