ATB Bremen has joined euroFMX, a major European research and innovation initiative developing trustworthy generative…
ATB’s VaMAI-Validator Research Receives Best Paper Award at ICAS 2026
How can developers determine whether an autonomous navigation system behaves safely, consistently and in accordance with maritime rules? ATB researchers addressed this question in the paper “VaMAI-Validator: Agent Validation Platform for Autonomous Maritime Navigation,” presented at ICAS 2026, the Twenty-Second International Conference on Autonomic and Autonomous Systems.
The conference took place from 8 to 12 March 2026 in Valencia, Spain. The paper, authored by Christoph Baier, Joel Aschmann, Feryal Fulya Horozal, Marcel Dechert and Sebastian Scholze, was presented by Christoph Baier. It was selected as one of the conference’s Best Papers based on the original submission, the camera-ready paper and the conference presentation. The award also includes an invitation to prepare an extended article for an IARIA journal.
A reusable environment for validating maritime navigation agents
Autonomous maritime navigation agents must make decisions in complex and potentially safety-critical situations. Their developers need to assess not only whether an agent reaches its destination, but also how it responds to other vessels, avoids collisions and complies with the International Regulations for Preventing Collisions at Sea, commonly known as COLREGs.
Such validation is often based on agent-specific simulation environments and manually prepared tests. This makes tests difficult to reproduce and results difficult to compare across different navigation systems. Reusing scenarios, assessment rules and performance metrics between projects also requires considerable development effort.
The VaMAI-Validator provides a modular environment for the simulation-based and automated validation of maritime navigation agents. It separates the navigation agent from the components used to evaluate it. Developers can define scenarios, rules and metrics as plugins and apply them to different agents. A network interface and client libraries support the connection of external navigation systems without requiring them to be developed as part of the Validator itself.
This architecture allows developers to establish a test-driven workflow for autonomous navigation: prepare the relevant scenarios and assessment criteria, execute the agent, analyse the results and use the findings to improve the next version. Because the same validation resources can be applied repeatedly, the platform also supports regression testing and comparisons between agent versions.
From test execution to detailed analysis
The platform evaluates navigation behaviour against defined rules and metrics. These can include collision-risk indicators, obstacle-avoidance requirements and simplified COLREGs checks. A graphical interface provides playback and detailed analysis of simulation results, helping developers trace problematic decisions and understand why a validation rule was violated.
The Validator can be deployed on premises and operated on standard laptop hardware. This supports work with systems or data that cannot be transferred to an external cloud environment.
For the study, the researchers integrated two navigation agents and developed three proof-of-concept plugins covering scenarios, metrics and rules. A case study involving a prototype path-planning agent showed that the platform could identify behavioural flaws during simulated navigation. The work therefore demonstrates how a common validation infrastructure can provide actionable feedback while reducing duplicated testing effort.
Further research will extend the range and realism of the scenarios, environmental conditions, vessel representations and rule implementations. Additional agent integrations will also be needed to assess the platform across a wider range of autonomous navigation approaches.
The paper was published in the ICAS 2026 proceedings by IARIA Press and is available through the open-access ThinkMind Digital Library.
Authors: Christoph Baier, Joel Aschmann, Feryal Fulya Horozal, Marcel Dechert and Sebastian Scholze
Further information:
Related Posts
