Description

With the increasing prevalence of mobile robots in human-centered environments, it has become particularly important to adjust their navigation behaviors through natural language. This research proposes a novel framework that uses Optimization Fabrics, a second-order dynamical system with provable stability, to achieve zero-shot, language-driven reactive navigation. By leveraging LLMs to directly transform user instructions into geometric structures and task-specific forcing potential fields, the study strictly separates the acceleration strategy from the priority metric. This mathematically ensures that user-specified behavior modifications can be executed in real-time while maintaining asymptotic stability, thus bridging the gap between human high-level semantic intentions and robust robot control.

The work is under the scope of the following projects:

• SOCIAL PIA: Cooperative Social PIA model for Cybernetics Avatars (Moonshot Research and Development Program) (web)
• TORNADO: foundaTion mOdels for Robots that haNdle smAll, soft and Deformable Objects (web)