The corona lockdown inspired a lecturer to design autonomous robots, which were a hit with students.
When the coronavirus caused the country to close down in spring 2020, a whole new way of teaching was opened up. At DTU Energy, a group of master’s and PhD students were meant to have been physically present on campus to learn how to manage an automated lab using machine-learning tools.
This wasn’t possible, since all students had to stay home because of the corona restrictions. But thanks to an unconventional solution, the classes continued. The lecturer, postdoc Rune Christensen, bought four sets of LEGO Mindstorms, which are normally used by children and young people to build robots. He installed an operating system in the robots and developed his own software, allowing him to use the machine learning tools he wanted. It worked surprisingly well.
The next step was to connect the students and robots on DTU’s network and use the University’s infrastructure. This made it possible to control the robots from home.
The solution created such a strong student involvement that Rune had to put in extra teaching hours because they wanted to work more with the solution to get the most valid data.
“The advantage of the robots is that we can give students a physical, experimental dimension without them being present. Even if you’re sitting at home, you can see the robot filling up containers with liquid, moving a conveyor belt, and measuring with a colour sensor. This makes the classes fun for the students—it’s something they want to do,” says Rune Christensen.
Mixing food colouring dissolved in water
Rune was inspired to design the robots by work that’s been done at the University of Toronto, which has extensive experience with automated laboratories. There, researchers had used a robot to mix coloured liquids, and that was exactly what Rune needed. But instead of buying an expensive robot, he developed new software and designed his own LEGO robots, which had originally been used to replace the experimental equipment that was packed away when DTU Energy moved from Risø to Lyngby.
The idea for the design came from a scientific journal where he found instructions for how to build LEGO robots that mix liquids.
Originally, the students were supposed to work with chemicals to produce materials for flow batteries. However, because of the risk that they might, for example, accidentally turn the lab temperature up too high while working from home, the setup was changed to mixing food colouring dissolved in water. The task was to program the robots so that they mixed green, red, yellow, and blue to achieve a certain orange colour. In this way, students could also learn to find strengths and weaknesses in the machine-learning tools.
“We considered running the classes using simulations, but agreed that it wasn’t the best way of learning. Even if you add random noise, simulations tend to work too well and too regularly. Students learn most when they gain experience with things that don’t work. There should preferably be some irregularities—and the robots can provide that,” says Rune Christensen.
Students wouldn’t accept poor data
To attend the classes, students logged into DTU’s computer system, which gives them access to program. The computer system was connected to the robots in a lab. The students were able to see the robots via webcams in the lab, where the lecturers took turns filling the containers with food colouring when needed.
The group work was done in interactive groups via the online tool Discord. Here the students had a virtual room where they moved around just like in a classroom. There was a central area where they could go up to the lecturers and get answers to their questions.
Lab classes with physical attendance have now resumed on campus, but Rune Christensen hopes the LEGO robots can inspire others:
“To be quite honest, you don’t exactly get perfect measurements using LEGO bricks. But I was surprised by how engaged the students were, and by their refusal to accept poor data. They got involved with the robots in a way they wouldn’t do with a simulation. It also rubbed off on those of us who ran the course.”