Heavy-duty mobile machines operate in harsh, high-risk environments such as forests, mines and construction sites, where reliability and safety are paramount. Any structural change or introduction of AI technologies to increase autonomy is therefore highly challenging in the field.
Mehdi Heydarishahna’s research tackles these transitions simultaneously, focusing on control systems, electrification and increasing autonomy. First, he proposes a method for controlling multibody machines that work with different energy sources, making it easier to switch from diesel-hydraulic systems to electric actuation. Second, he develops hierarchical control policies that incorporate learning-based components in a principled manner, leveraging their high performance while ensuring system safety in uncertain or faulty situations.
In his dissertation, Heydarishahna addresses the long-standing barriers of interpretability and trust in deploying black-box AI within safety-critical machines. By shaping the interaction between learning modules and robust controllers, his framework aligns with the emerging thinking in industrial safety and supports the gradual adoption of AI in real-world applications.
During his two and a half years as a doctoral student at Tampere University, Heydarishahna designed and validated the concept across multiple heavy-duty systems, including a hydraulic mobile robot, an electrified manipulator and a hybrid mobile robot – all in real time. Faced with diverse challenges of a similar underlying nature, he developed a generic control solution that is not only applicable when transitioning to electrified systems with minimal structural changes, but also adaptable to future energy sources.
“The energy source does not matter. We do not need to spend years forcing a full redesign whenever the technology changes. Our modular control makes upgrades simpler and independent of the energy source,” says Mehdi Heydarishahna.
According to him, we should not be afraid to use AI and learning-based methods in large-scale robots.
“We should deploy these high-performance, autonomous controllers and let them learn – even through mistakes and failures – until they succeed. We must design a supervisory module that can determine whether the AI-driven controller is behaving appropriately and then decide whether to let it continue learning and acting, or to switch to a non-AI controller. If you set the conditions wisely, AI will not disappoint you,” he adds.
Public defence on Tuesday 2 December
The doctoral dissertation of MSc (Tech) Mehdi Heydarishahna in the field of engineering sciences, titled Robust Deep Learning Control with Guaranteed Performance for Safe and Reliable Robotization in Heavy-Duty Machinery, will be publicly examined at the Faculty of Engineering and Natural Sciences at Tampere University at 13:15 o’clock on Tuesday 2 December 2025 at Hervanta campus, Konetalo building’s auditorium K1702 (Korkeakoulunkatu 6, Tampere).
The opponents will be Professor Alessandro Macchelli from the University of Bologna, Italy, and Professor Juha Röning from the University of Oulu, Finland. The Custos will be Professor Jouni Mattila from the Faculty of Engineering and Natural Sciences at Tampere University.