Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Posts
portfolio
publications
Enhancing Port Automation: A Novel Object Detection Pipeline for Container Ship Bays
Published in IEEE Sensors Conference, 2024
In port automation, efficiency and safety can be significantly increased by autonomous container handling of ship-to-shore cranes. A crucial aspect of achieving this autonomy involves accurately detecting common objects as well as the state of the bay. Previous research has primarily focused on detecting specific types of containers or hatches, which falls short in meeting the demands of complex automated operations. We propose a novel object detection pipeline specifically tailored for this application. In this pipeline the 3D point cloud data is initially transformed into a 2D representation, then fed into the Deformable DETR (Detection Transformer) model to detect objects of interest. Our pipeline successfully detects the positions and sizes of containers, hatch covers, open hatches, and bay areas. Notably, container detection achieves both precision and recall of 0.97. The mean absolute error of the container positions is smaller than 5 cm in all directions. The bay width can be predicted correctly in 99.8 % of the cases. These results are highly promising and pave the way for the automation of ship-to-shore cranes, leading to improved efficiency and enhanced safety.
Recommended citation: Junan Lin, Stefano Marano, Bruno Arsenali, Josip Marjanovic, Niklas Sundholm, Elin Jirskog, and Deran Maas*. Enhancing Port Automation: A Novel Object Detection Pipeline for Container Ship Bays. IEEE Sensors Conference, 2024.
Download Paper
Learning Over-Relaxation Policies for ADMM with Convergence Guarantees
Published in arXiv preprint, 2026
The Alternating Direction Method of Multipliers (ADMM) is a widely used method for structured convex optimization, and its practical performance depends strongly on the choice of penalty and relaxation parameters. Motivated by settings such as Model Predictive Control (MPC), where one repeatedly solves related optimization problems with fixed structure and changing parameter values, we propose learning online updates of the relaxation parameter to improve performance on problem classes of interest. This choice is computationally attractive in OSQP-like architectures, since adapting relaxation does not trigger the matrix refactorizations associated with penalty updates. We establish convergence guarantees for ADMM with time-varying penalty and relaxation parameters under mild assumptions, and show on benchmark quadratic programs that the resulting learned policies improve both iteration count and wall-clock time over baseline OSQP.
Recommended citation: Junan Lin, Paul J. Goulart, and Luca Furieri. Learning Over-Relaxation Policies for ADMM with Convergence Guarantees. arXiv preprint, 2026.
Download Paper