News & Events

Deploying Language-Guided Robots in the Wild

Harel Biggie, Postdoctoral Researcher in the Robust Robotics Group (RRG) at MIT

601 Computer Studies Building

To operate effectively on complex missions such as search and rescue or disaster recovery, autonomous robots must align their navigation with high-level natural language commands. This capability is essential both for enabling operation by non-expert users and for supporting the robots’ internal decision-making processes. These robots need more than geometric maps; they must reason over semantic features in ways that align with human mission descriptions. This talk presents work that leverages recent advances in foundation models to use natural language as a core representation for intelligent navigation. We introduce a framework that integrates classical path planning algorithms with code generation techniques for robot navigation. Combining the world knowledge embedded in Large Language Models (LLMs) and classical methods enables efficient, semantics-driven navigation towards a mission goal while providing a degree of explainability for robot actions. We further investigate how agentic frameworks integrated with graph databases can be leveraged to provide a shared understanding between a human operator and a robotic team. Components of this work have been successfully deployed in real-world robotic field demonstrations ranging from subterranean tunnels to large-scale forests. Finally, the talk will discuss practical considerations and lessons learned from deploying teams of robots in these harsh environments, including communication systems and platform design.

Bio: Harel Biggie is a Postdoctoral Researcher in the Robust Robotics Group (RRG) at MIT, advised by Professor Nick Roy. His research focuses on increasing situational awareness to improve decision- making in robot teams deployed in complex, unstructured environments by integrating natural language as a representation for semantics. Dr. Biggie completed his PhD at the University of Colorado Boulder in the Autonomous Robotics and Perception Group (ARPG), where he was advised by Professor Christoffer Heckman. At CU Boulder, Harel served as a project engineer and perception lead for Team MARBLE, CU Boulder’s entry into the DARPA Subterranean challenge for search and rescue robotics. His research directly contributed to the team's third-place place finish in the challenge. Before his graduate studies, he earned his Bachelor’s degree in Electrical and Computer Engineering from the University of Rochester.