RescueRoute AI

Inspiration During recent disasters, we noticed a recurring problem: most navigation systems are built for normal conditions, not emergencies. They assume roads are usable, data is reliable, and situations are predictable. In real disaster scenarios, none of this holds true. Roads may be flooded, blocked, too narrow for emergency vehicles, or overwhelmed by traffic and debris.

This gap between real-world emergencies and traditional routing systems inspired us to build RescueRoute AI. Our goal was to create a tool that acknowledges uncertainty and helps emergency responders make safer, more informed routing decisions when time is limited and lives are at stake.

What it does RescueRoute AI is an emergency routing and decision-support platform designed specifically for disaster response.

Instead of generating the shortest path, the system produces risk-aware routes that prioritize safety and reliability. Roads that are more likely to be flooded, congested, or operationally unstable are assigned higher costs, allowing responders to avoid routes that may fail in real-world conditions.

The platform also supports human-in-the-loop decision-making, enabling responders to inject local knowledge, block emerging threats, and adjust targets dynamically during live operations.

How we built it We built RescueRoute AI during the hackathon using a modular backend and an interactive geospatial frontend.

On the backend, city road networks are represented as graphs. Flood risk is modeled dynamically using a hybrid approach that combines elevation-based floodplains with runoff-driven flash flooding based on rainfall intensity. Instead of binary flood states, raster flood data is converted into node-level threat scores (0–100) that represent both direct flooding and surrounding instability zones.

Routing is performed using a risk-weighted cost function that balances distance, environmental threat, and road maneuverability. Narrow or residential roads are penalized more heavily than arterial roads, reflecting real emergency vehicle constraints. Manual threat inputs are also supported to allow operators to react to evolving conditions in real time.

The frontend visualizes flood extent, damaged roads, staging zones, and live routes, and is designed to support decision-making rather than fully automate it.

Risk-aware routing logic Traditional navigation systems treat roads as either open or closed. RescueRoute AI instead treats risk as a continuous factor, allowing routes to degrade gracefully as conditions worsen.

Each road segment’s cost is computed as:

Route Cost=Distance×(1+Flood Risk+Road Width Risk)

This allows the system to make realistic trade-offs. For example, a flooded highway may still be preferable to a flooded narrow alley, while a dry arterial road is strongly favored over residential streets.

Confidence score calculation To help responders understand route reliability, RescueRoute AI computes a confidence score for every generated route.

The confidence score is based on:

Average environmental threat along the route

Maneuverability risk from narrow road segments

Proximity to manually defined threat zones

A simplified representation of the scoring logic is:

Confidence=100−(Environmental Risk+Maneuverability Penalty+Threat Proximity Penalty)

Scores are intentionally capped to avoid false certainty, ensuring the system communicates risk transparently rather than overstating confidence.

Challenges we ran into One of the biggest challenges was modeling uncertainty without relying on heavy physical simulations, which are too slow and impractical for real-time emergency response. We had to carefully balance realism with performance so routes could update quickly while still reflecting real-world danger.

Another challenge was designing the system to remain useful even when data is incomplete or noisy, which is common during disaster situations.

What we learned Building for disaster response requires fundamentally different assumptions than everyday navigation. We learned that treating risk as a weighted cost rather than a strict rule leads to more flexible and realistic routing decisions.

We also learned that in high-stakes systems, clarity, reliability, and speed matter more than overly complex models.

Impact and future scope RescueRoute AI demonstrates how risk-aware routing can improve emergency response during disasters. In the future, the platform could integrate live sensor feeds, satellite imagery, and real-time incident reports to further improve adaptability and accuracy across different disaster scenarios.

Operational workflow (Demo explanation)

RescueRoute AI is designed around a phased emergency response workflow to reflect how real operations unfold.

Phase 1: Prediction The system simulates disaster impact based on rainfall input and visualizes flooded areas and damaged road segments. Flood risk is displayed as both water extent and operational instability zones.

Phase 2: Deployment (Ingress routing) Once conditions are assessed, the system scans for safe forward staging zones. These zones act as temporary operational bases closer to the target area. An ingress route is then generated from headquarters to the selected staging zone, prioritizing safety and accessibility under current conditions.

If no safe zones are found, the system automatically triggers Protocol Omega, enabling a direct but high-risk route from headquarters.

Phase 3: Live response (Rescue routing) From the staging area, the system computes a live rescue route to the target location. During this phase, operators can:

Move the target location dynamically

Manually mark emerging threats or blocked areas

Recalculate routes in real time

Each route is accompanied by an AI-generated confidence score to help operators assess reliability.

This workflow allows RescueRoute AI to function as a decision-support system rather than a static navigation tool.