A Lab for Self-Driving Vehicles

ETH Zurich students are learning about self-driving vehicles using a model with a fleet of mini-taxis. As part of the Duckietown course, they are working together with students in Montreal and Chicago on problems that concern developers of self-driving cars around the world.

A camera, an inexpensive mini-computer, a chassis, wheels and tiny LEDs are all it takes to build a Duckiebot. This is what Andrea Censi and Jacopo Tani – lecturers on the new course for self-driving cars at ETH Zurich – call these small robot taxis. Censi and Tani are senior assistants in the research group headed by Emilio Frazzoli, professor in the Department of Mechanical and Process Engineering and an internationally renowned specialist in autonomous systems.

self-driving vehicles
The duck is mascot only: In the Duckietown project, students taught the robots autonomous driving. (All pictures: ETH Zurich / Peter Rüegg)

On Tuesday, they presented their Duckietown teaching project in ETH Zurich’s main hall. Each of the small robot taxis has a trademark rubber duck on board. The vehicles navigate autonomously through Duckietown, a model town made up of streets marked out with tape, signs, traffic lights and parking spots. The duck taxis stop at intersections, yield the right-of-way and brake when they encounter traffic.

Problem-solving in teams

The students tinker with the hardware and software for days before it finally works. What looks simple is really a highly complex endeavour – so complex that the students are able to solve the tasks only through teamwork. They spend the semester working in groups on a certain part of the project. The course relies on global collaboration: Master’s students from ETH Zurich, the University of Montreal and the Toyota Technological Institute in Chicago are all working on the project simultaneously.

Censi developed the course at the Massachusetts Institute of Technology (MIT) at the end of 2015 in collaboration with Frazzoli and Tani and Liam Paull, now a professor at the University of Montréal. In 2017, they joined Frazzoli at ETH, where they continue to develop Duckietown.

Censi says: “We think it is very important to make the course a shared experience.” Team collaboration and management of a big project under time pressure are learning objectives – and skills in high demand in the industry. Nicolas Lanzetti, a Master’s student in Robotics and Control, says: “Normally, we study alone for our exams. Here, we have a common goal as a class: in the end, all the group projects must work together.”

Master’s student Sonja Brits adds: “We have had a lot of freedom in how we tackle the project and work together. But this also entails a lot of responsibility towards our fellow students.” Censi thinks this is the future of learning: “Theory and basic knowledge can be made available online. What is so valuable about a university like ETH Zurich is that it is a place where so many talented specialists can meet and collaborate.”

More complicated than in the books

The interaction of the various elements presents a big challenge in robotics and self-driving vehicles. Hardware components, sensors and motors must all be reconciled, both inside the Duckiebot itself – and among the fleet of self-driving vehicles. It is not the theoretical aspects that have perplexed the students, but rather the practical details: flawed hardware, components that do not interact properly, plans that do not work out. Censi says: “Thanks to Duckietown, the students recognise that things do not work exactly as described in the books. There are no perfect systems in robotics, and they learn how to deal with this fact.”

As the students teach the duck taxis to drive along the barrier lines smoothly, not be distracted by sunlight or to park cleanly, they are doing some pioneering work as well. The groups are all working on the same code. They document their solutions and make the information available for future students and researchers all over the world: the project embraces the open-source philosophy. Lesson modules, code and documentation are freely available online. Censi hopes this will inspire robotics enthusiasts from other institutions to develop their own self-driving Duckiebots to join the flock.

Out of the lab, on to the road

It is quite likely that knowledge and lines of code from the course will also be used to steer cars on real streets in the future. Emilio Frazzoli is co-founder of nuTonomy, which is planning a self-driving taxi service in Singapore. Andrea Censi is a systems architect at nuTonomy.

The students are also taking what they have learned and putting it to use out on the road: in 2017, the Akademischer Motorsportverein (AMZ) sent an electric self-driving racing car to the track for the first time – and won the Formula Student Driverless competition at the Hockenheimring. Several students from Censi’s course are already working on the AMZ driverless team and preparing for the next race.

Source : ETH Zurich