Computing, Engineering and the Built Environment students are encouraged to work on business-linked projects, aimed at developing digital and technology-based prototypes, addressing industry challenges, utilising the latest research, innovation and facilities. Final-year BSc Computer Games Technology student Marius Matulis, won first place at Innovation Fest for his project; the build, programming and 3D printing of a robot arm, trained by an AI digital twin.
Marius’ project, which uses the game engine Unity to virtually train a physical robotic arm, won first place at Innovation Fest, a three-day event offering businesses the opportunity to experience the creative talent emerging from Birmingham City University’s faculty of Computing, Engineering and the Built Environment. His ground-breaking work has since been published in Computers & Graphics, an industry leading journal.
We caught up with Marius to find out how he came to complete the innovative project:
"The foundations of the project began when I was in my first-year, where I won a prize at innovation fest for making three small dragons as a shooting game. This was my first experience of building and coding an electronics device, which I enjoyed. After finishing second-year, and just before starting my placement, I bought a 3D printer and started making a robot arm. It started as just a cool project - I never expected to develop artificial intelligence with it!"
"Unity, one of the main game developing real time engines, has improved and released Reinforcement Machine Learning (RML) toolkits. As a game developer, I saw an opportunity to use virtual environments to train a robot arm using RML."
"My professor, Dr Carlo Harvey, supported this idea from the start but warned me that it would be a huge project and task to complete."
Marius used reinforced learning simulations within the game engine Unity to train his physical robot arm in virtual space. Once trained, his arm was able to solve gripping problems using sensor input. By finding solutions to gripping problems, the virtual arm was able to feed the correct outputs to the physical robot for improved performance.
In the manufacturing process, a robot arm is typically pre-programmed with a set of actions and is unable to adapt to things going wrong. However, Marius’ virtual arm has been trained in a virtual environment, where things can go wrong and by observing the process millions of times, the virtual robot arm learns how to deal with problems that it has never encountered before unlike its physical counterpart.
"The robot arm was successfully trained virtually using the game engine, as the physical robot arm helped build a link between virtually trained artificial intelligence and the physical world. This concept is known as Digital Twin or Industry 4.0."
"My dissertation, “Robot Arm Training Using Reinforcement Learning in Fallible and Non-Deterministic Environments” has covered all the benefits and challenges of using this approach. There’s also been an academic paper published alongside this, it discusses possible future improvements and research areas related to the approach."
Project supervisor Dr Carlo Harvey said:
"This is a huge step forward from standard robot arms performing deterministic tasks into the world of non-determinism for manufacturing problems using game technology. This is a remarkable achievement for a final-year undergraduate student."
Marius added:
"Even though the course title is ‘Computer Games Technology’, these skills can be used in other areas, including: cross-country logistics simulations, VR/AR tool kits, human flow in train station simulations, construction virtual plan-out and many more."
"I am currently working in the automotive industry as a visualisation engineer and I use the game engine Unity to bring innovation to my work! My advice would be to always think outside the box – a game developer’s skill is not only for creation of games!"