Robust Deep Learning Control with Guaranteed Performance for Safe and Reliable Robotization in Heavy-Duty Machinery

Abstract

Today’s heavy-duty mobile machines (HDMMs) face two major transitions to advanced technologies: a shift from diesel-hydraulic to clean electric systems driven by climate goals, and a gradual move from human supervision toward greater autonomy. Firstly, diesel-powered hydraulic systems have long been the dominant actuation mechanism in HDMMs. Consequently, transitioning to fully electric systems, whether through direct replacement or complete system redesign, poses significant technological and economic challenges. Secondly, although advanced artificial intelligence (AI) technologies hold great promise for enabling higher levels of autonomy, their adoption in HDMMs remains limited due to stringent safety standards, and these machines still rely heavily on human supervision.

This dissertation aims to develop a novel control framework that: 1) reduces the complexity of control design for electrification of HDMMs by introducing a generic modular approach that is independent of the energy source and facilitates future system modifications; and 2) establishes hierarchical control policies that enable the partial integration of AI technologies into HDMMs while guaranteeing safety-defined performance and system stability.

To achieve this goal, five interrelated research questions (RQs) were formulated, which align with three overarching lines of investigation. The first line focuses on developing a generic robust control strategy for multi-body HDMMs that guarantees strong stability across different actuation types and energy sources. The second line seeks to design control solutions capable of maintaining strict predefined performance levels even under uncertainties and faults, while balancing the inherent trade-off between system robustness and responsiveness. The third line addresses the interpretability and trustworthiness of black-box, learning-based strategies, traditionally difficult to analyze and verify, toward enabling their stable integration into HDMMs in alignment with international safety standards.

The validity and generality of the proposed framework are demonstrated through three distinct case studies, involving different actuation mechanisms and operational conditions, and covering both heavy-duty mobile robotic systems and robotic manipulator systems. Collectively, the findings of this dissertation are documented in five peer-reviewed publications and one unpublished manuscript. This work advances the state of the art in nonlinear control and robotics, laying the foundation for accelerating the two aforementioned transitions in HDMMs.