Our summary
This project is a virtual laboratory that runs in a web browser. It lets people test how well an AI assistant can control a simulated robot arm. Users can set up tasks, like moving or stacking blocks, to see if the software makes the right choices.
The software asks the AI to look at the virtual scene and pick its next move. It can use Jev, an AI tool that chooses from options rather than writing an answer. The physics program then carries out the action and checks if the robot successfully grabbed the object.
This tool is useful for researchers comparing different AI assistants on the exact same task. The default demonstration uses basic rules and makes no AI calls. Connecting to cloud-based AI requires a private access code and costs money. Results only show what happened in these specific tests.
Key takeaways
- Start with basic rules to check the virtual physics and tasks before connecting an AI assistant.
- Compare different AI assistants using the exact same task, and match the video playback by simulation time.
- Save your test records, and do not record zero costs if the AI connection drops.