AI can make mistakes – and need to realise it

Researcher Chengyandan Shen is working on teaching artificial intelligence and robots to recognize when they are uncertain or making mistakes. While errors on a computer screen are usually harmless, mistakes by physical robots in the real world can happen on production lines, around expensive machinery, or inside hospitals.

Shen is a postdoc at SDU Robotics and participates in the MOTUS project, which focuses on monitoring and testing uncertainty and safety for smart collaborative robots. Through this research, her team collaborates with companies like Danfoss to automate real industrial setups and demonstrate that uncertainty-aware methods make robot behavior more reliable and efficient.

This research is part of the MOTUS project running from 2025 to 2027 with funding of 3.75 million Danish kroner. Project partners include the IT University of Copenhagen, the University of Southern Denmark, Universal Robots, Danfoss Drives, and ROPCA.

While the global artificial intelligence race often prioritizes size and speed, this work emphasizes the importance of reliability. Developing robots that know when they are wrong is a crucial step toward building trustworthy technology for factories and hospitals.