highlight
Design process
Desk research and document analysis
Game goal and game mechanism design
Game flow design
Interactive play test on paper
Low-fidelity prototype
Collaborate with Machine learning engineer to integrate AI agent
Train post-disaster responses with AI-powered game
Improve decision making in complex real-life scenarios
My role
Product designer
Team
Product designer
Game engineer
Machine learning engineer
INTRODUCTION
This AI-driven disaster response simulation game is designed to train American red cross employees in resource management decision-making under real-world constraints. By integrating a grid-based map system, AI agents, and reinforcement learning models, the game replicates the complexities of post-disaster scenarios—challenging players to allocate resources and adapt to unpredictable crisis scenarios.
CHALLENGES
This project was a true 0-to-1 design challenge, requiring constant iteration on play mechanism (through play test), AI agent, and resource dynamics, all while collaborating closely with engineers to ensure technical feasibility. Balancing realism and engagement was key to making the game both an impactful learning tool and an immersive simulation experience.
01
Overview
Why having this project and what we did?
When disaster strikes, every decision is critical. Through gamification, we hope to help disaster responders learn to make quick and effective decisions under pressure
Context
American Red Cross’s post-disaster training
Whenever a disaster strikes, ARC responds quickly to provide emergency assistance to affected individuals and families, including food, water, clothing and temporary shelter. Approximately nearly 70,000 relief operations are conducted each year, and about 90 percent of the staff are volunteers.
Tabletop Exercises are a common form of training for volunteers.
CONTEXT
Why American Red Cross want to digitalize boarding game tranining?
Advantages
of tabletop training
Train with collaboration between team members
Flexible rules and scenarios made by instructor
💡Keep
in digital game design
Disadvantages
of tabletop training
Limited capability to track decision data and analyze training performance, can’t contribute to future improvement
Lack of rigorous logic to simulate the dynamic event during post disaster management
Lack of real-time feedback system that help user make optimal decisions.
Inflexible training time and space constraints leading to low efficiency
💡Solve
in digital game design
Context
AI-social decision making lab: Researching how AI influence human decision
AI-SDM, lab collaborated with American Red Cross to form our team, aiming to research AI assistance to create situational awareness and decision support in disaster scenario, when human decision makers are cognitively burdened under time pressure and stress.
AI methods such as reinforcement learning and its variants, as well as cognitive models provide a workhorse for AI-enabled decision making, however validation of these methods fail to simultaneously capture the many complexities of real-world complex decision-making tasks such as uncertainty, partial observability, dynamics, heterogeneous temporal and spatial scales of decision making, polycentricity, limited communication, and ethics.
highlight
Design goal
Combining the needs from both ARC and AI-SDM lab, we aiming to design:
An AI-powered single-player game simulate Post-disaster resource management
Training need
Gamified strategy making training
Helping ARC employee become more proficient in disaster response
Research need
AI-enabled testing platform
helping researchers get valuable data and optimize AI supported decision making model
highlight
Game design principle
When designing the game, including goal, mechanism and interfaces, I follows four core design principles to ensure an effective, engaging, and realistic disaster response simulation:
Grid-Based Map System & Dynamic Event Generation
Creates realistic disaster scenarios with spatial constraints and evolving crisis conditions with realistic logic
AI Agents & Reinforcement Learning Models
Simulates real-world group decision-making complexity in single-player mode, providing multi-dimensional decision support.
Resource Allocation & Trade-Off Mechanism
Replicates real-world constraints by forcing players to navigate scarcity and risk in disaster response decision making.
highlight
Our solution
02
Game mechanism
How can we achieve our design goals through the design of the gameplay?
Design player goal
Who is the player? What is their goal?
Our first step in determining what we need to determine is player's ultimate goal in the game, which is strongly related to our training purpose
Through research, we found that ARC is not very sensitive to cost control in the actual process, the core of ARC is people-centered and humanitarian, the concept of customer first and life first.
Player: Experienced volunteers that need to learn coordination and high-level management strategy
Achieve high client satisfaction score
Users need to focus on response time, resource allocation efficiency, and task completion rate, which are all strongly correlated with user satisfaction.
Resource distribution ROI
Optimize Budget spent per resolved case or community and minimizing waste (e.g., unused food or supplies).
Design mechanism
Game flow
Around this play goal, we developed the initial game flow to structure the game phases and task execution process. The game spans 8 days.
Before the game, the instructor sets disaster parameters like affected areas, population, weather, and event probabilities.
On Day 1, players build shelters and kitchens, assign volunteers to prepare for following game process. Throughout the game, daily, they constantly adjust strategies to meet survivor needs such as lodging, food, and return-to-home settlement.
Emergencies like road collapses, vehicle failures, and extreme weather create disruptions, demanding quick trade-off decisions under pressure.
Detailed game flow of everyday resource management
User’s Jobs-to-be-done
-
Infrastructure Planning: Build shelters and kitchens based on predicted demand.
-
Volunteer Assignment: Deploy volunteers to shelters, kitchens, and logistics with optimal efficiency.
-
Food Distribution: Deliver meals via ERVs, balancing speed and satisfaction.
-
Assign displaced survivors to shelters or motels.
-
Adjust resources based on sudden events.
AI agents dynamically engage players, providing insights, constraints, and feedback to help them navigate complex trade-offs and evolving scenarios.
Refine game play flow with interactive prototype
Key thing to research
Design mechanism
Planning functionalities in game interface
With insight from user testing, combined with defined game goal and user flow, I designed the layout of the game interface with understanding of key functionality modules.
Follows the principle of minimal disruption by integrating navigation and key functions into the main screen. Essential controls are embedded in the environment, avoiding deep hierarchies and keeping critical actions easily accessible. This allows users to stay focused on their primary task without unnecessary UI friction.
Final UI
Design mechanism
Key mechanism 1: Grid-Based Map System & Dynamic Event Generation
Problem
How can the game simulate the complexity and unpredictability of real-world disasters and the pressure they create?
Research insight
Structured unpredictability balances logic and uncertainty over pure randomness, help to train adaptive decision-making in disaster scenarios
Volume of daily task demand and Possibility of emergency events are dynamically triggered by environmental factors and grid attributes, simulating real-world crisis unpredictability. Here is the behind-stage logic I designed for event generation
Example
Design mechanism
Key mechanism 2: AI Agents & Reinforcement Learning Models
Problem
How can the game realistically simulate ARC employees’ teamwork and decision-making processes?
Research insight
Effective disaster response relies on collaboration—AI in single-player mode must act as human-like teammates
Conversational AI with Personality & Bias
AI agents act as distinct team members with unique perspectives and limited scope of information
Information Aggregation & Decision Assistance
Reduces cognitive load by summarizing key insights for resource allocation.
Explainable Feedback
Helps players understand the reasoning behind decisions
Provide Adaptive Decision Scoring
RL tracks player and agent choices over time, using a reward function to measure decision
Game State Monitoring
RL collects real-time player context, including
-
Available resources
-
Disaster status
-
Task performance
-
Player decision patterns
Example
Provide user feedback and information
Make initial decision and suggestions, let user refine the decision
Design mechanism
Key mechanism 3: Resource Allocation & Trade-Off Mechanism
Problem
How can players effectively develop resource allocation strategies through gameplay?
Research insight
Resource scarcity in disaster response forces competing priorities, making trade-offs and priority evaluation the most critical skills
I designed the resource system and trade-off mechanism refer to game strategy and real life situation, here are the five core resource in the game, when making a decision respond to an event, players must balance at least two competing resources. Impact of decision also may delayed. Overcommitting today may waste resources tomorrow, training players to think long-term and strategically.
Example
when a nearby kitchen fails to meet food demand, players must weigh:
-
Option 1: Spending more money on fast food for immediate relief
-
Option 2: Transporting food from a distant kitchen, risking lower satisfaction and exposure to future crises
-
Option 3: Building a new kitchen, investing in long-term stability but risking resource misallocation.
Example
when an emergency disrupts food delivery, players must choose between:
-
Option 1: Reallocating from another kitchen – Saves money but delays delivery, lowering survivor satisfaction.
-
Option 2: Purchasing fast food – A costly but immediate fix to maintain satisfaction.
-
Option 3: Paying for repairs – A moderate trade-off, sacrificing some money and time for long-term efficiency.
03
Game interface
How do you make a game design that enhances the player's gaming experience?
Design interface
Optimize the main page
Original prototype
Feedback from user test
Design strategy
Highlight and prioritize tasks: Action oriented
Re-Grouping elements according to functionality
Adding zoom-in/out feature and make map full-screen
Design interface
Optimize the task pop-up
Original prototype
Feedback from user test
Design strategy
Integration of AI agent in processing tasks
Emphasize context and scenario to support decision making
Highlight estimated impact
Design interface
Optimize the task pop-up
Original prototype
Feedback from user test
Context-aware information convey
Conversation on current and previous task decision making
Design strategy
Integrated into tasks and context
Immersive conversation
Easy access on main page to historical decisions and new conversation
✌🏻Let's get in touch!
+1 4124039378
04
Reflection
What I learned and What is next?
My takeaways
-
Tech meets real needs – Explored multiple AI technology to address real-world challenges in decision-making, learn the mechanism and pros/cons behind the technology.
-
Reality-inspired, playability-first – Every mechanic was designed with game experience in mind to ensure engagement.
-
Beyond traditional UX – Designing a system driven by nonlinear flows and game mechanics required flexible playtesting and a shift from screen-based UX to interaction-based game design.
-
Psychology-informed design – We grounded key decision points in behavioral research and literature, ensuring every trade-off felt realistic, meaningful, and cognitively engaging.
Next steps
-
Refine user experience, interface and interaction with support of user research, making it more intuitive and engaging
-
Collaborate with MLE to research AI-agent integration and model training strategy
-
Support game engineer to develop the game in Unity
-
Go beyond MVP: start to design other dynamic scenarios and research on NLP scenario generation