IEEE CASE 2019 Tutorial on
Theory and Education of Autonomous Control of Driving Agents with Deep Learning
Organizers: Amazon Artificial Intelligence Lab., Seattle, WA, USA; University of California, Berkeley, CA; Mila, Université de Montréal, Montréal, Canada, ETH Zürich, Zürich, Switzerland; nuTonomy, an Aptiv company, Santa Monica, CA, USA; University of Washington, Seattle, WA, USA
Goal: Deep learning resulted in achieving human-level capabilities in computer vision applications such as face recognition and machine vision in self-driving cars. However, the integration of these new advancements in sensing theory and applications into control theory education has been lacking. To close this gap and identify the needs and requirements in education and industry in a realistic environment, we are organizing a tutorial using AWS DeepRacer (https://aws.amazon.com/deepracer/). In this tutorial, you will train the AWS DeepRacer virtual car on the race tracks in the cloud-based 3D racing simulator, and for a real-world experience, you will deploy your trained models onto 1:18 scale AWS DeepRacer and race your colleagues on the real racetrack.
Part I – Lectures on advance sensing and controls (8:30-10:00 am)
Lecture 1: The Artificial Intelligence Driving Olympics (AI-DO)
Lecture 2: Autonomous Vehicles: An Open Platform for Learning and Teaching
Lecture 3: Bayesian Policy Optimization for Model Uncertainty
Coffee break (10-10:30 am)
Part II - Hands-on learning with AWS RoboMaker & Sagemaker: Model training on cloud (10:30 am-11:30pm)
Lunch break (11:30 am-1:30pm, boxed lunch will be provided)
Part III - Hands-on racing with AWS DeepRacer: Model validation and comparison to the traditional approaches (1:30-3:30 pm)
Coffee break (3:30-4:00 am)
Part III - continued (4-5pm)
Requirements: The participants are not required to have any prior knowledge on deep reinforcement learning. Basic knowledge in embedded and real-time controls and machine learning are beneficial. The participants will be provided with Amazon Cloud credits during the tutorial.