Overview & result
YuanKJing shows Jev-as-Policy, built on Jev: One-click simulation setup with Jev as the control policy.
From the original post
Jev-as-Policy by @yuankjing: One-click simulation setup with Jev as the control policy.
How Jev fits in the loop
- A local harness turns MuJoCo geometry and contact state into compact text and JSON.
- A first TypeSafe call asks Jev for the intent.
- A second call asks for motor choices as labels, for example x negative, y positive, z stay, fingers close.
- The harness maps the labels to a Cartesian servo with local IK, and MuJoCo moves the Franka Panda arm.
How to reproduce
- Review the source documentation and repository at https://github.com/YuanKJing/Jev-as-Policy.
- Configure environment with TypeSafe Jev / System One API keys.
- Run test prompts against the decision endpoints.
Why this build matters
Shows Jev in practical use for robotics, IoT and computer vision.
Limitations
- It is a reproduction of the public "Jev as Policy" control structure on one demo task.
- Jev receives no images and returns no joint angles, coordinates or trajectories.
- Each policy update needs two dependent TypeSafe calls.