21700. kzfaq.info/get/bejne/r8yjntKSnMjDoIE.html

Still working on it here and there. Focusing on the machine-learning part of the project using reinforcement-learning to teach AI to play foosball from self-play. Code here if you're interested: github.com/glmcdona/FoosRL Created a self-play virtual foosball table in Unity, and using RL to teach it to play foosball. I've only had very limited success so far in teaching it to play very effectively, but I'm still plunking away at it here and there. The robotics hasn't progressed at all, since I've realized that with current state of the art RL methods to learn from self-play you need years of play experience to train one version of a model. To do this effectively you need the virtual environment so it can learn by being able to simulate play much quicker than a real world table automated with robotics.

@@glmcdona21 Thank you for the code. I'm interested in the project. I saw some some universities and companies starting projects in the field. Bosch: kzfaq.info/get/bejne/iLmJpsuCycnOiYE.html EPFL: kzfaq.info/get/bejne/Y9iYi9thybClYoE.html What I find very interesting is that it is a closed physical system. So you could train the AI against itself for days (with a mechanism to bring back the ball) and see what kind of style it develops. Like if it learns to pin the ball, use the edges, etc.