Embodied Active Learning Playground

This demo is inspired by the works on general embodied active learning. There is a hidden rectangle beneath this environment, and you can use your mouse to move a sensor to collect binary measurements. Each time, the sensor measures up to a certain number of dots within its field of view. A measurement point (dot) is white if it is outside the underlying rectangle, and it is black if it is inside the rectangle. We have an optimal “generative model” that samples predicted rectangles (red boxes) using the current measurements as the dataset. Your goal is to move the sensor to collect data for the model to generate more accurate predictions of the underlying rectangle. Try it out with different sensor configurations!