Going for Robotic Gold

Going by the unforgettable name of Duckietown, the project is an interdisciplinary learning exercise in robotics research, giving students experience prototyping self-driving robots and applying artificial intelligence (AI) to the physical education platform developed by Massachusetts Institute of Technology (MIT) for experiential learning. And while the Duckiebot vehicles may have been AI-driven, the overall project was very much student-driven.

A group of engineering students first learned about it while participating in a hackathon organised by Law, Innovation, Technovation and Entrepreneurship (LITE) Lab@HKU. One of LITE’s founders, Mr Brian Tang from the Faculty of Law, told them about Duckietown, and they took the idea to Dr Loretta YK Choi, Faculty of Engineering for advisory and technical support, and to help them raise knowledge exchange funding.

Dr Choi said: “Our protocol for this was that students would take the lead and enjoy as much flexibility as possible while working towards the project goals, so that they could explore, make mistakes, assume responsibility, self-adjust and practise autonomous learning to grow better and stronger, both personally and intellectually.”

The eventual team was made up of students with knowledge and skills from computer science, mechanical engineering and electronics engineering. Later students from HKU’s Artificial Intelligence Reading Group also joined.

Their task was to prepare robots for competitions, in which they would perform autonomous real-life tasks, such as lane-following, on a physical platform. In the run-up, the teams underwent intensive training, fostering peer collaborations, solidifying their knowledge and demonstrating sustainable learning outcomes.

From the fun point of view, the project highlight was the chance to compete in the grand finals of the international AI Driving Olympics (AI-DO), in May 2019, co-run with the International Conference on Robotics and Automation (ICRA).

“While there they were also able to meet peer contestants with common interests from other academic institutions, such as MIT, ETH Zürich, Université de Montréal and National Chiao Tung University, and to exchange ideas and knowledge with them,” said Dr Choi.

The heart of the AI-DO was Duckietown, a physical miniature replica of a driving environment, where each Duckiebot [vehicle] has a single camera attached to it. Students experimented with different AI algorithms on a simulator and devised their own in the context of an impactful auto-driving problem. It enabled them to actually see how their solutions work in a playful, physical environment.

At the ‘Olympics’, they also learned the true meaning of Steve Jobs’ words: “When we arrived at the AI-DO contest in Montreal, the AI model the team had developed was still not working very well,” said Dr Choi. “Team members tried very hard to improve the model the day before the live challenge for which the submitted AI programs were deployed to the physical Duckiebot for action in Duckietown.

Accidentally off-road

“We were a bit nervous when it came to our turn to run our algorithm on the Duckiebots, but it turned not too bad, and in fact it was hilarious when our Duckiebot drifted away and went out of bounds when negotiating a turn in the road!”

While their Duckiebot did not win, it was second runner-up, a creditable performance for a first try, and the experience as a whole gave the students invaluable experience in being team players solving problems on the spot.

“The project also reinforced and consolidated the skills and knowledge that students learned from their respective disciplines,” said Dr Choi. “They researched areas of machine learning and auto-driving and got to experiment with advanced technology.”

Team member Kelvin Ng, a student from the Bachelor of Engineering in Mechanical Engineering, said: “In Montreal, not only did we gain practical experience in hands-on techniques for the driverless car project, we also had a better taste of applying imitation learning to typical control engineering challenges in the mechanical engineering discipline.”

Fellow team member Angel Woo, a second-year student from the Bachelor of Engineering in Computer Science programme, said: “I was once a ‘software’ person who was uninterested in hardware… however, this project helped me learn the fun part about hardware, and experience how software and hardware can be integrated in reality.”

While this first venture into AI-DO was very successful, Dr Choi hopes that in 2020 they may be able to expand the scope even further.

“At the start, we had envisioned that while determining the algorithmic parameters for autonomous driving, students would also need to deal with some behavioural control especially in a complex, multi-vehicle environment,” she said. “But it turned out that vehicle control by AI is by itself difficult to master within the time span (students had less than half a year to work from scratch), so in the end the team was not able to explore these other aspects.”

However, this has not deterred Dr Choi, nor the students – more are already signing up in 2020. Indeed, such is the enthusiasm that they are also exploring whether it is possible for them to host their own contest within HKU so more students can participate.

“It has been an incredibly rewarding experience to work with our undergraduate engineering students who were passionate and self-motivated in delving into the evolving AI technologies and applications,” said Dr Choi, adding that a large part of the appeal of Duckietown was the constant reminder that learning can and should be fun, and a little foolish.