Please Note : Due to confidentiality reasons, I cannot share the bulk of my work. Additionally, certain images/ deliverables have been obscured to preserve confidentiality. Thanks for your understanding!
PS. For a deeper understanding into the project, check out the thesis work here - Human-Centered MAS Thesis

Context

ABB's Research Group was interested in exploring the potential of Multi-Agent Systems (MAS) as an emerging approach to address the growing complexity in industrial operations. They recognized that MAS could help tackle longstanding challenges, such as fragmented workflows, coordination breakdowns, and cognitive overload - by enabling intelligent, distributed decision-making. This project was initiated to understand how MAS could be designed in a human-centered way to better support operators and improve trust in automation.

Work Context of Target Group (Voyage & Technical Operators)

Background

To ground the research in a real-world setting, I focused on ABB’s Fleet Support Service Center, which provides continuous operational support for global vessel fleets. This context was selected because its dynamic and distributed nature, with operators collaborating across technical and voyage domains, which offered a rich environment to study how human-agent collaboration unfolds in high-stakes scenarios. The target groups studied were:

Technical Operators – specialists monitoring and analyzing ship/vessel equipment health and system diagnostics.
Voyage Operators – experts coordinating ship/vessel routing and operational decisions in real time.
Management Stakeholders – leaders shaping strategic priorities, policies, and expectations for system use.

Role & Responsibilites

UX researcher & designer creating the end-to-end experience of this technology, with activities including:

Deep literature/secondary research
Field Studies utilizing Contextual Inquiry
In-Depth Interviews
Thematic Analysis
Creating guidelines for Human-Centered MAS
Designing & prototyped interfaces to support the guidelines
Documentation, Demonstration and Handoffs

Duration & Team

Duration: 6 months (Jan 2025 - June 2025)

The project was solely my responsibility & therefore, every task listed is done by me. Supervisor of the project helped with setting up contacts for the research part of the project.

Outcome

1) A comprehensive set of human-centered design guidelines for Multi-Agent Systems in industrial settings that offer a practical framework that bridges gaps between computational capabilities and human-centric requirements, contributing to improved efficiency, reduced cognitive overload, and enhanced operator trust and overall adoption of such technology.

Guidelines for Human-Centered Multi-Agent System

2) Interaction concepts and interface patterns demonstrating how these principles could be applied in practice, including:

Action Explanation Panels
Mixed-Initiative Controls
Shared Mental Model Dashboards
Voice-Activated Commands
Predictive Forecast Overlays
and much more!

Example of an interface design pattern of Transparency & Explainability mechanism

Approach & Process Overview

Planning

I began by clearly defining the research objectives, identifying key questions, and setting the project scope. Through iterative discussions and reviews, I aligned the goals closely with ABB’s strategic interest in exploring Multi-Agent Systems (MAS). This phase also involved understanding

Previous Research within ABB: Reviewed existing internal studies and initiatives to build on ABB’s prior exploration of AI and automation.
Identifying Potential Research Context: Evaluated different operational domains to select a setting where MAS could deliver the most impact, and ABB’s Fleet Support Service Center came out as the most relevant context.

Research

Research Aim & Questions

To identify and approach the research pragmatically and holistically by understanding the global and local landscape, I framed the Primary research question as - How can a multi-agent system be designed to be human-centered and support coordinated decision-making within industrial contexts?
The sub-questions to help answer this were:

What are the critical needs and challenges of operators that can be addressed by MAS interactions?
How could a multi-agent system address these needs in a human-centered way?
How could the identified design factors support coordinated decision-making in fleet operation support services?

a) Secondary Research

To ground the project in evidence and avoid reinventing the wheel, I conducted deep literature and secondary research on previous MAS implementations—analyzing what worked, what failed, and why. This involved looking at over 100+ peer reviewed research studies on Multi-Agent Systems, Human-Agent Interactions, Human-Centered AI, Industrial Automation etc.

b) Contextual Inquiry

To build an authentic understanding of real-world practices & working context I carried out contextual inquiries with Voyage and Technical Operators, observing workflows, collaboration patterns, and pain points in situ. Each contextual inquiry lasted ~2.5 hours and furthermore, during the observations, the following 4 factors were looked out for to get an even better & holistic view of their workflow & tasks:

Questions & Issues - Obstacles or blockers that users face.
Tools/Artefacts - What helps the person get their task(s) done
Knowledge dependent - Is there something that they needed to know or should have known, for the task/process to be successfully completed.
Workarounds - Was there something that they utilize that is unintended, that helped them with their work

c) In-Depth Interview

To get an even better and true understanding of the work ecosystem, it was important to figure out what guides their work and how strategies and policies are prioritized and governed, and therefore I conducted in-depth interview with management stakeholder to explore strategic goals, perceptions of automation, and expectations for future systems. The interview was semi-structured and lasted ~60 mins.

Analyzing & Synthesizing

Thematic Analysis

To ensure the findings would lead to actionable insights, I applied thematic analysis to all qualitative data, systematically coding field notes and interview transcripts to identify recurring challenges, needs, and opportunities, which resulted in 15 bigger key themes (with several subthemes) to be generated.

Zoomed out due to confidentiality reasons

Journey Mapping

To make sense of complex workflows and highlight pain points and specify opportunity areas while making sure that future teammates understand the researched workflow, I mapped operator journeys and visualized interactions, revealing the real-world friction points and communication gaps and highlighted opportunity areas for Multi-Agent System to come into picture!