HiveWay
Inspiration
I've lost count of how many times I've gotten lost looking for a room in Georgia Tech's Klaus Advanced Computing Building. It's one of my favorite buildings on campus, but its hallways can feel like a maze. I started wondering what would happen if a fire blocked the only way out I knew.
In the 2000 Boland Hall fire at Seton Hall University, two survivors recalled trying to escape through thick smoke with a stairwell nearby that they rarely used. Their experience made the question behind HiveWay feel urgent: How can we help students find another way out when their usual route is blocked, even if they barely know the building?
What it does
Open HiveWay, select your floor, and it estimates your location inside Georgia Tech's Klaus Building. It then shows the shortest available route to an exit, taking into account which corridors and exits are blocked in the fire simulation.
How we built it
1. Indoor Map Digitization
We started by visiting every floor of Klaus Building and taking photos of the floor plans posted inside. Since these were simple floor guides rather than detailed digital blueprints, we had to manually reconstruct the hallways, rooms, and exits as SVG vector maps, keeping the scale consistent across floors.
But having a visual map wasn't enough for navigation. We needed a way for our algorithm to understand which hallways were connected and how far someone would need to walk. To do this, we placed nodes at regular intervals along the center of each corridor and connected neighboring nodes with edges, creating an undirected graph that represents the building's walkable paths.
The next challenge was connecting this custom map to the real world. GPS provides latitude and longitude, but our SVG maps use their own local coordinate system. We used Google Maps to identify the building's geographic position and align our digital maps with its real-world footprint. We then implemented a coordinate transformation that converts GPS coordinates into positions on our indoor maps, allowing us to place the user's estimated location directly onto the floor plan.
2. Shortest-Path Routing
Once we had a navigable graph, we needed to determine which exit a user should take. Rather than assuming the nearest exit was always the best option, we used Dijkstra's algorithm to calculate the shortest path from the user's position to every exit in the building. We then compared the resulting distances or estimated travel times and selected the closest reachable exit.
However, a shortest path is only useful if it's actually accessible. To account for the fire simulation, we designed our routing algorithm to exclude nodes blocked by fire. This way, the algorithm searches only through available corridors and avoids recommending routes that pass through blocked areas.
3. Dynamic Fire Simulation and Rerouting
Finding a safe-to-traverse route once wasn't enough. In a fire, conditions can change while someone is still evacuating. A corridor that was open when navigation began might become blocked just moments later.
To handle this, we connected the fire simulation directly to our navigation graph. As the simulated fire spreads, additional nodes become unavailable. Whenever this happens, HiveWay runs Dijkstra's algorithm again from the user's updated position to every available exit, comparing the remaining routes and selecting the shortest reachable one.
We also built a navigation view that updates the user's displayed position as they move along the route. Together, these features allow HiveWay to adapt the recommended path to both the user's progress and the changing fire conditions, including switching to a different exit when necessary.
4. Interactive Visualization
Finally, we needed to make all of this understandable to someone navigating an unfamiliar building. We built a browser-based application using React and TypeScript, using our custom SVG maps as the visual foundation.
On top of each floor map, we overlay the graph's edges, highlight the recommended evacuation path, and display the user's position and simulated fire locations. When a route changes, the highlighted path updates on the map so users can immediately see where to go next.
By bringing the digital floor plans, pathfinding algorithm, fire simulation, and navigation interface together, we turned our manually constructed building graph into an interactive evacuation prototype.
Challenges we ran into
1. Building an Indoor Map Without Reliable Floor Plan Data
Our first challenge came before we could even begin implementing the routing algorithm. Unlike outdoor navigation, where services like Google Maps provide detailed road networks, we couldn't find a digital indoor map or API for Klaus Building. Even the floor plans posted inside the building were simplified guides without precise measurements.
This meant we couldn't simply import an existing map and start calculating routes. Instead, we used the photographed floor plans as a starting point and manually reconstructed the building's layout at a consistent scale. We then translated the hallways into a connected graph, making sure the connections reflected where people could actually walk. By building our own spatial data from these limited references, we were able to create a usable foundation for indoor pathfinding.
2. Dealing with Unreliable Indoor GPS
Our biggest technical challenge was figuring out where the user actually was inside the building. Although we had developed a way to translate GPS coordinates onto our custom maps, the GPS readings themselves were often too inaccurate to identify the correct hallway.
At first, we tried matching each GPS reading to the nearest node in our navigation graph. But even small location errors could cause the displayed position to jump into an adjacent corridor or suggest that the user had moved through a wall.
We tried several approaches to improve positioning accuracy, including movement constraints to reject physically impossible jumps. However, none of these corrections could overcome the fundamental inaccuracies in the indoor GPS readings.
This led us to try pedestrian dead reckoning (PDR) as an alternative to continuous GPS tracking. Starting from a known position, we used the phone's orientation to detect changes in walking direction, particularly turns at hallway intersections. We also analyzed the phone's motion to estimate step counts and stride lengths, using the number and intensity of detected movements to approximate how far the user had walked. By combining estimated distance and direction, we could update the user's position along the indoor navigation graph without relying on every new GPS reading.
Built With
- gps
- react
- sensorzoo
- svg
- typescript
- vite
