ScentUSV Automated test scenario synthesis for verifying collision avoidance of autonomous vessels
Abstract
Recent years have seen massive efforts into developing unmanned surface vehicles (USVs) for both civilian use (e.g., oceanography, cargo transport) and military applications (e.g., surveillance, transport, minesweeping, search-and-rescue, and combat). In such contexts, a USV often needs to complete its mission autonomously under varying environmental conditions (e.g., weather, traffic restrictions) while not endangering other maritime traffic. To avoid collisions between ships, the International Maritime Organization (IMO) has formulated the International Regulations for Preventing Collisions at Sea (COLREGs), a set of international rules describing how a vessel should navigate in close-encounter situations with other vessels. Therefore, COLREGs compliance is also crucial for the safe operation of autonomous USVs in the presence of other maritime traffic. However, since the COLREGs rules are formulated with human operators in mind, implementing them as part of the control system for autonomous vehicles is nontrivial. While COLREGs compliance has been demonstrated for USVs in both simulations and controlled field trials, most existing efforts have been limited to simplecollision avoidance scenarios. Currently, however, there is a lack of effective techniques for assuring COLREGs compliance of USVs in complex traffic scenarios, involving multiple vessels and/or static obstacles. Such complex scenarios can represent extremely rare combinations of events and special circumstances, which are unlikely to be covered by traditional (stochastic) simulations. Ongoing research on safety testing for autonomous road vehicles heavily relies on automatically synthesizing challenging test scenarios which are then investigated in a virtual setting in sophisticated traffic simulators. This way, potential safety concerns (e.g., near-crash situations) can be revealed without jeopardizing human life. The ScentUSV project aims to adapt similar principles for system-level safety assurance of USVs. Compared to previous efforts in maritime test scenario synthesis, a key novelty of our proposed approach is the use of model-based test generation employing qualitative abstractions. This has the dual benefit of (1) drastically improving scalability by avoiding spending time on test cases belonging to the same equivalence class, and (2) providing safety assurance by enabling formal coverage and diversity guarantees.The expected outcome of the ScentUSV project include (1) a set of abstract models of maritime actors and collision-avoidance behavior, with an accompanying test scenario specification language; (2) new techniques and methods for systematic, scalable generation of complex test scenarios in the maritime domain; (3) practical integration of developed test-scenario generators with existing naval simulators; and (4) new methods for evaluating situation coverage and diversity in generated test scenarios. These results will be disseminated via publications in top scientific conferences and journals of software and systems engineering.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 15, 2023
- Source ID
- N629092412006
Entities
People
- Daniel Varro
Organizations
- Linköping University
- Office of Naval Research
- United States Navy