What it is
A browser-based MuJoCo Franka Panda workbench for embodied AI experiments. It offers three simulation tasks, supports rule baselines, local MiniCPM5-2B, TypeSafe Jev, Claude, and OpenAI-compatible APIs, and visualizes each decision step.
How it uses Jev
Jev can choose from bounded simulator actions over structured observations. Deterministic code owns physics and safety checks. The README describes Jev as a decision model entry, but does not specify the exact state fields or how Jev's returned probabilities are used in code.
Primitives:choice
Technique worth stealing
Merge independent questions under the same state and directly output finite candidate probabilities to reduce repeated input and text generation.
Try it
Run 'embodied-jev serve --port 8090', open http://127.0.0.1:8090, select a task and model, then run the experiment.
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricno0.28
- Shows a System One patternunclear0.53
- Handles uncertaintyno0.06
- Measuredno0.07
- Runnableyes0.85
- Worth recommendingunclear0.40
- Model replicano0.13
- Problem scopescore on a 0–2 scale0.69
- About Jevyes0.95
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.