As robots become increasingly capable, many real-world tasks require multiple robots with complementary skills to work together over extended periods of time. Enabling such teams to understand high-level human instructions and translating them into coordinated actions remains a fundamental challenge.

Our research investigates how large language models can enable intelligent task planning for heterogeneous robot teams. We combine the language understanding and reasoning capabilities of LLMs with structured representations, symbolic planning, and environmental feedback to develop robot teams that can determine what needs to be done, which robot should do it, how actions should be coordinated, and how plans should adapt when the situation changes.

A central focus of our work is bridging the gap between high-level language and reliable physical execution. Rather than treating robots, tasks, and environments as isolated components, we develop planning approaches that reason about robot capabilities, object relationships, spatial constraints, and dependencies between actions. This allows teams of robots to efficiently coordinate complex, long-horizon tasks while reducing planning errors and unnecessary computation.

Ultimately, we aim to enable adaptive multi-robot systems that can understand, plan, coordinate, and act in complex real-world environments from natural-language instructions.