Projects

Research projects

Current and previous projects that translate IMPACT LAB research into intelligent engineering design, manufacturing, robotics, and autonomous systems.

Our projects bring together AI, computing technologies, engineering modeling, and human-centered workflows to support design, analysis, manufacturing, robotic operations, and autonomous engineering systems.

Ongoing projects

AI-Integrated Design

AIDIO - AI-Integrated Design, Inspection, and Optimization

AIDIO develops an AI-agent-based platform for ship design, inspection, and optimization. The project addresses a central bottleneck in automated shipbuilding: the lack of a continuous digital thread from design intent to robot-ready production data. By coordinating design agents, rule-grounded inspection, multi-objective optimization, production simulation, and governance-enabled execution, AIDIO aims to help shipyards generate consistent design artifacts, verify them against engineering and classification constraints, and refine them through closed-loop feedback before physical execution. The project treats AI as part of an integrated engineering workflow rather than a standalone drafting assistant, connecting specifications, geometric models, bills of materials, process plans, and change histories. This architecture supports safer automation by making design decisions traceable, inspection results explainable, and production data suitable for downstream robotic execution in complex shipyard environments.

AIDIO workflow diagram linking ship design, AI agents, inspection, optimization, simulation, and robot-ready production data.

Agentic Systems Engineering

SEMAA - Systems Engineering Modeling with Agentic AI

SEMAA develops an intelligent modeling assistant for model-based systems engineering in maritime and ocean engineering. Built around recent advances in large language models, agent frameworks, and domain-grounded knowledge organization, SEMAA is designed to understand engineering commands, interpret domain-specific language, and generate useful SysML and systems engineering artifacts. The project aims to bridge theoretical MBSE concepts and practical industrial workflows by integrating modeling tools, structured knowledge, and expert feedback from real-world engineering collaborators. Its modular architecture is intended to connect SysML modeling tools, external databases, engineering documents, and AI reasoning services into a practical ecosystem. Through collaboration with industry partners, SEMAA also studies how agentic AI can capture domain knowledge, reduce manual modeling effort, and accelerate the creation of consistent systems models for increasingly complex maritime engineering projects.

SEMAA architecture diagram showing AI-assisted systems engineering modeling and tool integration.

Process Intelligence

AIPM - AI-Integrated Process Management

AIPM creates an intelligent, model-based platform for managing complex engineering and manufacturing processes in the switchboard domain. The project makes process knowledge explicit by representing tasks, dependencies, roles, resources, versions, and change impacts in forms that both people and AI systems can use. By combining process modeling, scheduling, version control, GraphRAG-grounded reasoning, and human-in-the-loop decision support, AIPM aims to reduce fragmented manual work, improve visibility into risks and status, and support better planning and coordination across design and manufacturing activities. The platform links tools such as Visual Paradigm, PlantUML, Microsoft Project, GitLab, and knowledge retrieval services into an operational toolchain so process models can become executable planning information. AIPM also emphasizes change management, version awareness, and impact analysis, helping teams understand how design changes propagate through manufacturing tasks, resource assignments, schedules, and project risks.

AIPM process management architecture connecting modeling, scheduling, GitLab, GraphRAG, Microsoft Project, and AI agents.

Self-Organizing Manufacturing

SmartSOM - Self-Organizing Manufacturing System Integration

SmartSOM investigates self-organizing manufacturing systems for personalized production in volatile markets. The project develops multiagent learning methods that allow distributed manufacturing agents to coordinate without relying on rigid centralized control. By combining task-constraint modeling, social learning, organizational learning, multiagent reinforcement learning, and hardware-based validation, SmartSOM aims to create manufacturing systems that can adapt to changing product recipes, uncertain demand, and heterogeneous resources while remaining robust, efficient, and scalable. The research extends ideas from cybernetics, cognitive learning, and organizational learning into computational frameworks for team design and system adaptation. It also develops metrics for task complexity, robustness, adaptability, and learning efficiency so decentralized manufacturing systems can be evaluated against centralized and heuristic baselines. Hardware-in-the-loop validation is used to study how simulation results transfer into real manufacturing environments.

SmartSOM conceptual diagram showing multiagent learning and self-organizing manufacturing systems.

Robotic Operations

VLAG - VLA with Graph Modeling for Robotic Operations

VLAG develops a modular vision-language-action framework for long-horizon robotic operations. Instead of relying on a single monolithic policy, the project uses graph-based task routing to coordinate specialized compact experts for perception, language understanding, planning, and action generation. This structure is intended to make robotic intelligence more interpretable, scalable, and efficient for manufacturing jobshop environments where robots must understand instructions, perceive changing workspaces, and execute complex sequences of actions autonomously. The approach decomposes robotic intelligence into task-relevant experts and routes among them through graph representations of goals, actions, and dependencies. By separating routing, perception, language grounding, and action execution, VLAG explores a path toward smaller and more transparent robotic models that can support flexible manipulation without requiring a very large foundation model to reason through every step.

VLAG architecture diagram showing graph-based routing for vision-language-action robotic experts.

Previous projects

Autonomous Ships

AutonoShip - Building a Technological Basis of Autonomous Ships

AutonoShip developed a technological foundation for autonomous ships through the ShipWare framework and a set of intelligent applications for navigation, collision avoidance, engine-room monitoring, remote autonomy, and regulatory support. The project explored ROS-based ship system architecture, multi-layer sensing and decision-making, and AI approaches to collision avoidance from mathematical, human-knowledge, and reinforcement-learning perspectives. The work contributed to autonomous vessel decision support by connecting situation awareness, risk modeling, human steering data, and simulation-trained intelligence within a unified ship autonomy framework. The system concept organized autonomy around bridge operations, engine-room autonomy, remote control assistance, and legal or regulatory reasoning. Its ShipWare architecture connected sensing, localization, situation awareness, guidance, and control, while the AI intelligence framework examined multiple ways to generate safe decisions under uncertainty. Together, these elements supported research on risk-aware navigation, autonomous collision avoidance, and intelligent monitoring for future autonomous maritime systems.

Project video Press release
AutonoShip system framework diagram showing ShipWare and autonomous ship application layers.

More will be added.