We enable autonomous robot teams to learn both how to act and how to collaborate—deciding which agents should coordinate, with whom, when, and under what resource, uncertainty, and reliability constraints. To this end, we pioneer Value-of-Coordination metrics and mission-grounded distributed optimization methods for cooperative missions of combinatorial complexity, with applications to intelligence, surveillance, target acquisition, and reconnaissance (ISTAR).

Our approach goes beyond conventional ergodic search and consensus mechanisms for distributed action coordination and scene understanding. Key thrusts include: (i) mission-oriented distributed optimization, including self-organized consensus for scalable action coordination and sensor fusion, guided by Value-of-Coordination metrics pioneered in our lab; and (ii) risk- and resource-aware behavioral adaptation and decision-making driven by mission-grounded large language models and world models.

Our morphable quadrotor, MorphQuad, acts as a flying human hand—to transform agriculture, emergency response, infrastructure inspection and maintenance, and beyond. Each rotor uses an independently controlled two-axis gimbal, allowing maximum thrust in any direction while retaining a compact, miniaturizable structure. Its controller addresses gimbal lock, control singularities, and rotor downwash interference. Using fully onboard sensing and computation, MorphQuad demonstrates multi-revolution pipe inspection, valve turning, wall perching, pushing and pressing with human-wrist-level strength, and stable recovery from wind and physical disturbances.