Improving Tesla’s self-driving AI using virtual worlds

This is a great little video from the Two Minute Papers channel. There’s a lot going on inside the video, so I’d like to first add a tiny bit of context …

Firstly: one of the many reasons that Artificial Intelligence is improving so rapidly, is that the research community is very open in developing and sharing their research results in the form of academic ‘papers’. Anyone in the world can point their browser at https://arxiv.org/list/cs.AI/recent (a popular online service for academic papers) and browse the very latest in AI research results as they are shared with the world!

Secondly: these are academic papers, so don’t feel too bad if you don’t understand what they are saying! And for this video, I recommend you let your eyes glaze over with the technical details and focus on the ‘gist’ of what’s been saying. However, if you do one of the many online courses in AI (I recommend Andrew Ng’s deep learning specialization at Coursera, alternatively the YouTube playlist with the course videos) to understand the key concepts and lingo, you might be pleasantly surprised at how you will quickly become able to kind of follow along with the explanations in many cutting edge research papers!

Thirdly: Two minute Papers is an excellent channel to keep abreast of the most recent developments in AI research. Dr Károly Zsolnai-Fehér does all the hard work of filtering through the firehose of academic research to present entertaining and informative highlights of the especially interesting or impactful research.

Finally: Tesla has made a big bet on camera-based AI for their self-driving efforts. Unlike Google’s Waymo or Uber’s (recently sold) self-driving developments where radar and lidar are heavily used to map the surrounding environment, Tesla relies exclusively on a set of cameras looking in all directions, from which they reconstruct the size, distance and speed of all objects, as well as constructing an overhead view of the area – a ridiculously impressive feat using purely camera pixels! Look out for it in the video …

Cool! With that bit of context out the way, jump in and be amazed 😎