Robot-Use LLM Agents
Robot-Use LLM Agent Harnesses
Custom harnesses that make an existing LLM a better robot-use agent.
The model is only part of a robot-use agent. The harness around it decides what the model sees, what it can call, and how its output becomes motion. We research harnesses that let an existing LLM perform better on a robot without retraining it.
The end goal is for a small model, such as qwen-3.8-27b or qwen-3.8-next-flash, to control a robot well, and for models such as gpt-6-astra and haiku-5.5 to operate machines better than they do out of the box.
Branches
Manipulation
VLM-driven manipulation.
Navigation and Locomotion
Moving through the open, real world.
World Models
Giving the model a working picture of the world it acts in.
Small Models
Getting more robot capability out of small, low-cost models.