CHOOSEAgents & automation
Jev Model for Robotic Simulation
Zhuo Tao@runzhuotao
1:29This project utilizes the Jev model to interpret sensor data from a robotic simulation, generate action options, and select the optimal command for task execution.
TakeJev selects categorical options containing numerical commands; it does not generate unrestricted coordinates or joint rotations.
Ok I tried the Jev model for a robotic simulation.
Observe: MuJoCo provides depth points/ASCII maps, object locations, joint positions, velocities, actuator loads, finger contact and slip measurements.
Interpret in parallel: separate Jev calls assess the task, geometry and contact/dynamics. Noul judges conditions, Score assesses priorities and stability, and Choice selects intent and focus.
Generate options: code samples numerical base destinations, gripper positions, compact-arm poses and jaw commands from the current geometry. Collision and reachability checks screen them.
Compare in parallel: Jev evaluates several candidate groups. Their top alternatives go to a final call, which selects the complete action, speed and grip force.
Execute and repeat: deterministic code performs inverse kinematics, turn-then-drive base motion and smooth trajectories. New observations, completed targets and stalled motions feed the next decision.
Jev selects categorical options containing numerical commands; it does not generate unrestricted coordinates or joint rotations.
The experiment required navigating two barriers, approaching a raised table, and grasping and lifting a cup. And the robot passed in one trial.
