Building My First Mixed Fleet Warehouse Simulation: From Workshop Sticky Notes to a Fully Playable Training Environment

Over the past months, I’ve been quietly working on something that started as a simple co-design workshop idea, a few sketches, and a pile of sticky notes from the Mixed Fleet participants who identified the most challenging human-AGV interaction moments in warehouses. Those early discussions eventually became a fully playable simulation built in Unity, which still feels slightly surreal to admit. And just to clarify, I am not a programmer. This was my first real dive into C#, physics scripting, object orientation, and debugging. The fact that the simulation now runs, turns, signals, and behaves in a way that resembles real warehouse movement is something I’m genuinely proud of.

The simulation’s three scenarios were directly shaped by our co-design workshop insights. The first scenario, the narrow corridor bottleneck, places the participant at the end of a tight corridor with an AGV already working inside it. Participants must decide whether to wait, backtrack, or boldly attempt to squeeze past the machine, revealing how they evaluate familiar trade-offs between risk, efficiency, and trust. The second scenario, the intersection task, throws participants into a crossroads at the exact same moment as an AGV, forcing them to interpret its light cues and movement patterns and decide who has the right of way. It tests their situational awareness and their ability to “read” the behavior of an automated teammate. The third scenario, the deadlock, simulates the moment an AGV suddenly stops in the participant’s path, no warning, no explanation. This moment captures how people react when automation behaves unpredictably, and how quickly trust can shift. 

Building all of this in Unity was a chaotic but oddly satisfying process. I combined my own warehouse models with ready-made Unity assets, and then painstakingly wrote C# scripts to control AGV behavior, navigation logic, vehicle physics, and signaling. There were days when I broke everything before lunch and fixed everything after dinner. I learned how to make vehicles accelerate realistically, how to synchronize visual feedback with behavior, and how to stop AGVs from breakdancing in circles (which they enthusiastically tried to do whenever they could). It was messy, frustrating, and honestly one of the most rewarding things I’ve done.

Small peak into the building of the simulation. From left to right: 1. view from a forklift driver’s seat; 2. C# coding snippet of AGVs lights; 3. view of the warehouse aisles.

And then came Wednesday, 19.11, a small milestone that felt huge. We officially finished collecting data with logistics students and teachers from TREDU. Watching them navigate the simulation was eye-opening in the best possible way. They were fast, reactive, brutally honest, and, most importantly, they exposed every design flaw I didn’t even know I had. Their feedback wasn’t just “useful”; it was essential. They helped us identify what needs improvement, which cues were confusing, which behaviors looked wrong, and where the simulation failed to communicate the AGV’s intent clearly enough. At the same time, they provided rich data for our research questions on trust, awareness, and decision-making. Some of their responses surprised us, some confirmed what industry workers had previously said, and some pointed us toward new directions entirely. 

TREDU students performing tasks in the simulated mixed fleet environment.

That session reminded me that building the simulation was only half the work, the other half is letting real users stress-test it, challenge assumptions, and show us where the “logic” of the system falls short of the messy reality of human behavior. 

This simulation isn’t just a technical achievement. It is a glimpse into what future training and human-robot research tools may look like: interactive, grounded in real industry needs, and capable of revealing subtle but critical moments of trust and miscommunication. It provides a safe and replicable way to study situations that would be too risky or disruptive to recreate on a warehouse floor. And because it’s built directly on the insights from our Mixed Fleet workshop and validated by actual future logistics workers, it captures real challenges instead of hypothetical ones. 

For the Mixed Fleet project, this simulation represents a novel step forward in exploring human-robot teamwork. For me, it’s proof that with determination (and a heroic amount of Googling), even someone who never planned to code can create a complex, meaningful research tool. And honestly, this feels like only the beginning of what simulations in logistics training and trust research can become. 

Written by Veronika

The blog was originally published at Tampere University web site.