Fresh is not the same as right: decision-aware update scheduling for maritime digital twins, tested on real AIS, radar and LiDAR data
Year: 2026
Type: preprint
Abstract: A maritime digital twin often cannot take every update it is offered, because links, compute or operators are limited. Which updates should it take first? We compare four policies under equal update budgets (round-robin, maximum age of information, maximum predicted uncertainty, and a decision-critical policy that favours uncertain vessels near the collision-alert boundary) and score them by how well the twin's CPA/TCPA collision alerts agree with a twin that receives every update. All data are real: AIS for San Francisco Bay and New York Harbor, shore radar with AIS in the port of Hamburg, and LiDAR, radar and camera recordings from small vessels in Trondheim and the Pohang canal, about 550,000 runs in total. In a pre-registered test on held-out New York data, the decision-critical policy raised collision-alert recall by 0.06–0.19 over the freshness policy at budgets of 5–50% of offered updates, and roughly doubled precision (0.77 against 0.36), while its twin was the stalest of the four; at every budget the freshest twin was not the one that decided best. A predicted growth of the gain with traffic density was not confirmed. In exploratory tests on Hamburg shore radar, recall improved by 0.05–0.11. In harbour recordings with one to five craft, recall differed little across policies. These results concern agreement with full-update alerts; the evaluation did not measure collision outcomes.
Author: Mamun, S M Abdulla Al
Institution: University of Manchester
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7236723834
