The First AI Model in Longboarding?

Student Halenur Öztürk posted an absolute highlight internally today – her first AI model capable of recognising the movements involved in a longboard trick, a frontside no-comply. In our longboard and skateboard laboratory at the GECKO Institute at Heilbronn University, we have recorded several thousand videos of longboard dancing and freestyle sequences. Some of these tricks and moves were recorded in large numbers to enable the effective training of AI models. Halenur had videos converted into skeleton data using an (AI-based) model and then trained a BiLSTM model with this data. There are still a few kinks to sort out, but the success of her final thesis is now a foregone conclusion ;-) Nice!

Why is this significant?

Well, we are interested in teaching humanoid robots to longboard. No machine can do that yet. There isn’t even a humanoid robot that could carry a rucksack for us on a mountain hike. They cannot (yet) master the skills of rolling, sliding and gliding. But that would be very helpful for the practical applicability of these machines, which are likely to remain always expensive. Yet, it is still a long way to go in our journey. But, the team has gone more than just a few steps now.