CATCH

Research

Research

How the work is structured, where it runs, what is underway, and what has been published.

How the work is structured

The work is decomposed into five design echelons. Each holds one or more instances, emits an artifact the next echelon takes as input, and iterates internally while validating coherence with its neighbours.

Design knowledgeaccumulationProblemanalysisProblem statementObjectives andrequirementsDesign requirementsDesign anddevelopmentSolution designDemonstrationProof of conceptEvaluationProof of value
  • Design echelon type
  • Echelon instance
  • Intermediate artifact
  • Macro-level design process
  • Micro-level iteration within an echelon
  • Coherency validation between echelons

The structure is the eDSR metamodel of Tuunanen et al., 2024. Each echelon holds one or more instances and emits an artifact the next echelon takes as input. Select an echelon to read it.

  • LUT

01Problem analysis

Establish a validated problem statement through systematic investigation of coordination failures.

Instances

  • Coordination failure analysis

Activities

Literature review across emergency management, human–AI collaboration and interaction analytics. Structured interviews with rescue service coordinators. Analysis of after-action reports from recent Finnish operations. Focus groups with emergency management professionals.

Artifact and knowledge output

Problem statement. A validated problem statement specifying when current AI systems fail, which coordination breakdowns occur, and what cognitive mechanisms underlie them.

AI and HCI Behavioral Lab

A behavioural laboratory at LUT University for studying how people interact with systems, from intelligent systems to other digital environments. It is equipped to record what happens while someone is using something, with eye trackers, functional near-infrared spectroscopy (fNIRS), GSR and other physiological sensors, and VR/AR.

Studies run here produce the empirical measurements CATCH works from, and contribute to state of the art research across several fields.

AI and HCI Behavioral Lab at LUT University

Illustrative traces of the signal types recorded in the lab. Not recorded data.

Currently underway

  1. Reading cognitive state from interaction

    A review of what is known about inferring cognitive and functional state from ordinary interaction behaviour: mouse movement, the timing of actions, and how often someone returns to information they have already seen. This is the foundation the rest of the project stands on. If those signals carry enough about attention, workload and fatigue, adaptive support becomes possible in settings where you cannot attach sensors to someone mid-emergency.

  2. An instrumented emergency operations game

    A small multiplayer game placing players inside an emergency operations centre as dispatcher, field coordinator and resource manager, coordinating under time pressure while the scenario escalates and scripted disruptions arrive. It is instrumented throughout: every interaction and every spoken exchange is logged, which produces realistic and repeatable moments of overload to study, and two signal streams to test inference against.

  3. Interviews with emergency operators

    Interviews with operators to establish where coordination actually breaks down, which decisions are genuinely hard, and where an operator would want a system to intervene. Literature and simulation only reach so far; these are questions only practitioners can answer.

Publications

2026

  • Jan-Paul Huttner, Dominik Siemon

    Nascent Design Principles for Human-Machine Interfaces in Airborne Manned-Unmanned Teaming

    DESRIST 2026, Design for Better Futures: Beyond the Science of the Artificial. Lecture Notes in Computer Science, volume 16607, pages 42 to 54conference

    Abstract

    Manned-unmanned teaming (MUM-T) design remains theoretically fragmented despite growing operational deployment across civilian and military aviation domains. This research-in-progress addresses this gap through a literature-grounded design science study embedded in the echelon design science research framework. Based on a systematic synthesis of 10 fixed-wing MUM-T empirical studies, we derive four theory-grounded, nascent design principles: Context-Adaptive Automation Intervention, Hierarchical Delegation Granularity, Integrated Temporal-Spatial Plan Visualization, and Bidirectional Human-Autonomy Negotiation. The contribution is positioned as Level 1 nascent design theory, consolidating fragmented findings into actionable prescriptions justified by cognitive engineering and human-automation interaction theories. The principles establish an initial prescriptive knowledge base for subsequent artifact instantiation and evaluation with operational pilots.

    DOI

Open science

How the project releases what it produces. Links appear here as each output is published.

Preregistration

Confirmatory hypotheses are preregistered on the Open Science Framework before any data is collected, together with the study design, the planned sample size and its power analysis, the analysis plan, and the success criteria. This is what keeps hypotheses from being written once the results are already known.

Open code

The interaction analytics toolkit is released as a Python library on GitHub under the MIT licence, documented for people outside the project who want to use it on their own data.

Open data

Anonymised datasets covering interaction logs, physiological measures and performance outcomes are published on the Open Science Framework, so the analyses can be checked and reused.

Open access

Papers are published open access through green or gold routes, so nothing produced here sits behind a paywall.

Complete reporting

Null results are reported with the same rigour as positive ones, with effect sizes and confidence intervals, full model specifications, sensitivity analyses, and a plain account of the limitations.