Presented Jev-as-Policy framework for managing Franka Panda robots

Developers have published Jev-as-Policy — a specialized control structure for the MuJoCo simulator. The tool is designed to automate manipulative tasks of robotic systems, allowing the conversion of text descriptions of states into specific actions. The system uses TypeSafe API to select manipulation intentions and determine the Cartesian coordinates of goals. In the current implementation, the framework provides basic operations for the Franka Panda robot, including grasping, lifting, and moving objects such as blocks in a virtual environment. The solution is focused on integrating language models with low-level motion controllers. The use of typed API calls minimizes errors when transmitting commands from an AI agent to a physical or simulated manipulator, ensuring predictability of trajectories during task execution. The project is available in an open repository and provides a ready-made base for testing control algorithms in MuJoCo. The development is aimed at simplifying robot programming through natural language, eliminating the need to write complex scripts for each individual grasping or moving operation.