RescueTwin

Inspiration-

Our inspiration came from the tragic disaster that occurred in Brooklyn, New York, on Wednesday, November 11, 2026. The incident showed how delayed communication and emergency response can cost lives. Some of the harm may have been prevented if people had received immediate guidance and first responders had access to accurate, real-time information. This inspired us to create a digital twin bot that simulates disasters, guides people toward safety, and helps first responders prioritize life-saving actions.

What It Does

Our project is a digital twin bot that creates real-time simulations of disasters and emergency situations. It analyzes the surrounding environment, identifies possible dangers, and gives people clear instructions for reaching safety. At the same time, it relays critical information—such as hazards, safe routes, and people who may need immediate help—to first responders. By improving communication and decision-making, our project aims to reduce response times, prevent confusion, and save lives.

How We Built It

We built our project by combining an accurate fire simulation, native pathfinding, and an AI-powered emergency assistant. The simulation models how a fire spreads through the environment and how conditions change over time. Native pathfinding identifies safe evacuation routes and redirects people when fire or debris blocks an existing path. The bot converts this data into clear, step-by-step instructions, while a first-responder dashboard displays hazards, safe routes, and the locations of people who need immediate help.

Challenges We Ran Into

One of the biggest challenges was balancing between realistic physics simulations (fire/smoke and gas+sprinkler system) while maintaining real-time performance. Another big challenge was developing the best algorithm for robot navigation/path planning that could adapt when fire, smoke, or debris blocked an evacuation route. We've implemented a mission-level A* search for survivor extraction and integrated an agentic reasoning architecture using Gemini. The process of trial and error was long but we managed to identify many clear and reliable extraction methods during emergencies. Finally, we had to define the clean data contract and provide smooth API integration for both the physics engine and autonomous robot to communicate with each other, follow rules and behavioral limits so the bot would prioritize safety, avoid misleading instructions, and respond appropriately in different disaster scenarios.

Accomplishments That We’re Proud Of

We are proud that we successfully simulated several important parts of an emergency, including realistic fire and smoke behavior, responsive sprinkler systems, and native pathfinding that guides people around hazards. We also integrated complex 3D data, including OBJ files and detailed mesh renders, to create a more realistic environment. Bringing these systems together allowed us to build an immersive disaster simulation that demonstrates how technology can improve emergency preparation and response.

What We Learned

We learned how to utilize large-scale datasets and programmatically produce high-quality synthetic data ingestion and real-time pipelines for performance-heavy software. We learned how to deal with messy data and identify the right sources to use and implement workarounds when the ideal method is failing. For example, a realistic viewer for the simulation using Matterport was impractical with the given API and datasets, as the Matterport SDK doesn't provide full high-res scanes/splats of interiors, so our workaround was identifying scenes with mesh and materials available through Matterpak.

What’s Next for RescueTwin

Our next step is to expand RescueTwin beyond fire emergencies by adding simulations for floods, earthquakes, severe storms, and other disasters. We also plan to fine-tune the inference model for each scenario so the bot can provide more accurate and situation-specific guidance. Future improvements could allow the bot to navigate multiple floors, coordinate evacuation routes across an entire building, and eventually operate through a physical robot that can assist people and first responders in real-world emergencies.