12:30h - 13:20h
Designing the Brain of your Autonomous Car
Self-driving cars are closer than ever before, but we still need to do better in order to reach fully autonomous vehicles. Do you want to help us shaping the 'brains' for the next generation of cars? It's actually very easy (and free!) to do your own experiments, you will see ;) Short Description: This talk targets both beginners that don't know a thing about Artificial Intelligence (AI) or Reinforcement Learning (RL) and would like to know about how these new cars are going to work, and experts in RL and Deep RL (DRL) that want to know about more advanced techniques (such as Inverse RL and Formal Methods). If you would like to make a self-driving vehicle but don’t know how to start without breaking your own car... This is your talk! Extended Description: In this talk we will dive into Deep Reinforcement Learning, the field that is the closest to reach the ultimate 'brain' for our cars and how can we experiment with simulated cars by ourselves. We'll start with an intro about AI and DRL, the AI technique that achieved some of the most impressive breakthroughs in the last few years: https://www.youtube.com/watch?v=8dMFJpEGNLQ ; https://www.youtube.com/watch?v=UuhECwm31dM
However, why do we still struggle with Autonomous cars? To understand this we will cover the following - Some very basic math concepts such as Markov Decision Processes and Bellman equations. - What are policy-based and value-based algorithms and how they are pruned to different driving styles... from a Formula 1 to a safety paranoid! - Good practices when designing "rewards" for our AI so that it doesn't behave like a gluttonous lazy child neither it commit suicide (Yep, they actually try to kill themselves if you aren't careful...) - What is failing with current DRL and more advanced techniques, such as Inverse Reinforcement Learning and Formal Methods (with DRL), for those of you who want to go further than DRL! - Open-source books, car simulators... to start your own AI path!
Borja González León
PhD Candidate and Graduate Teaching Assistant
Imperial College London
Researcher in AI: Improving Generalization skills in Multi-agent Deep Reinforcement Learning with Temporal Logic. Previously focused on applied research with both Satellite Images and Autonomous Cars.