The team runs the first live demo of Monty on real-world data! It’s a very special moment. They show that models trained in simulation can be recognized in the real world despite lighting conditions and all sorts of noise. The demo was developed during a one-week hackathon.
Note: After the hackathon week, we cleaned up the code and fixed a few bugs, which means that now we actually get better performance than what we report in the video. We also created a set of benchmark experiments to test different conditions (bright, dark, occlusions,…).