Autonomous Vehicle Olympics Competition Hopes to Spur Innovation

TU-Editor

---

14th August 2018

A competition to develop the best in artificial intelligence (AI), technology for self-driving vehicles is sponsored by the nonprofit Duckietown Foundation.

The Duckietown Foundation, a nonprofit organization promoting robotics and AI, announced the first AI Driving Olympics (AI-DO), a competition to develop AI for self-driving cars. The competition comprises four key challenges of increasing complexity that autonomous vehicles (AVs) will need to master to achieve higher levels of autonomy on real-world roads. These include road-following on an empty road, road-following with obstacles, point-to-point navigation in a city network with other vehicles, and fleet-planning for a fully autonomous mobility-on-demand system.

The competition will use the Duckietown platform, a miniature self-driving-car platform used for autonomy education and research. Competitors will have access to a suite of simulators, logs and baseline implementations, while real environments called Robotariums will be remotely accessible for evaluation. Contestants can also build or acquire their own testing facility (Duckiebots and Duckietowns) either through open-source DIY instructions or as rewards obtained in the organization’s Kickstarter campaign.

The foundation counts among its backers some heavy hitters in the AV space, including the Toyota Technological Institute at Chicago, nuTonomy (an Aptiv company) and Amazon Web Services. According to the announced timeline, the next two months are set for the open development of the competition and include periodic releases of simulators, tools and baseline implementations.

October first marks the official opening of the initial AI Driving Olympiad Robotarium, with live environments available for use. The highest-scoring entries in the Robotariums will be run during the live final event in December at the Neural Information Processing Systems (NIPS) machine learning conference in Montréal.

“The AI Driving Olympics offer a glimpse of the challenges of creating self-driving cars,” Emilio Frazzoli, CTO and chief scientist at nuTonomy, and senior advisor at Duckietown, said in a statement. “A playful but rigorous competition on a smaller and safer testbed is the best way to develop the creativity needed to make progress in this field.” Duckietown and AI-DO are developed at ETH Zürich, among other locations, including Tsinghua University in Beijing, National Chiao Tung University in Taiwan and Georgia Tech.

The second edition of the AI Driving Olympics, is already scheduled for next May, in conjunction with the International Conference on Robotics and Automation (ICRA) in Montréal. Autonomous driving is a multifaceted engineering challenge but, among other things, it’s a massive object-recognition problem. Even small changes, like putting bits of tape on a stop sign, can throw off an automated recognition system, so the software needs to be resilient enough to recognize such things.

Organizations across the globe are pouring money into the research and development of autonomous vehicle systems, which require complex, large-scale test facilities. The Toyota Research Institute (TRI) announced in May that it is constructing a closed-course test facility where it will develop AV technology.

The Michigan site expands the institute’s closed-course testing capabilities, adding to partnerships with GoMentum Station, in California, and Mcity and the American Center for Mobility, both located in Michigan.

— Nathan Eddy is a filmmaker and freelance journalist based in Berlin. Follow him on Twitter.