Inspiration
This project was inspired by a keen interest in search problems, integrating emerging technology, and the outdoors! We chose an "in-the-clouds" ambitious project by tackling a task at the intersection of decision making with uncertainty, robotics, and search and rescue.
What it does
LIFELINE is an autonomous system for drones to find lost hikers. Using a prior estimation of the hiker's possible locations, several drones distribute a search area to find the hiker as fast as possible. The outcome is a dramatic decrease in rescue times which directly translates to lives saved.
How we built it
We used a mixture of web programming stacks (React, Express) and techniques in efficient computation with Python libraries like NumPy. Algorithms were implemented in Python leveraging libraries for optimized delivery with results communicated via IPC. A node.js WebSocket server provides by-frame updates about drone location and position. A comprehensive React interface created an immersive, real-time dashboard exploring how this system integrates into a real commercial entity.
Challenges we ran into
Building a project as challenging as this, with difficulties on multiple fronts, was quite the struggle. We had to overcome computational problems with sending large state spaces, solving specific optimization problems fast, and creating a cohesive narrative that described a very fast-moving project.
Accomplishments that we're proud of
We made some pretty good progress for 24 hours! One thing we're especially proud of is that LIFELINE works end to end. A missing-person report becomes a probability map of where the hiker is likely to be. A fleet of drones then splits up the search area through an auction, sweeps the most likely areas first, and updates its beliefs every time it fails to spot the hiker. When a drone reports a possible sighting, a ground team is sent in to confirm the find.
Our model of where a hiker might wander is grounded in Lost Person Behavior research, combining Koester's distance rings with the terrain and elevation of the park. It also accounts for tree cover. Dense forest makes drone cameras more likely to miss someone, so a search that comes up empty over heavy canopy lowers the odds less than one over open ground. The Python engine streams every tick through a Node WebSocket server into a live React command center. The result feels like a tool a park ranger could actually use: launch the drones, review possible sightings, dispatch a ground team, and confirm the find.
What we learned
We learned how much of search and rescue is really about reasoning under uncertainty. A sweep that finds nothing still tells you something, and putting that into numbers changes where you should look next. We also learned how hard it is to coordinate many agents in real time. Drones have limited batteries and flight times, and they must not duplicate each other's work. On the engineering side, we learned to split heavy numerical work in Python and NumPy from real-time delivery through ZeroMQ, Node, and WebSockets, and to keep the messages between them small enough to stream smoothly. Finally, we learned that a strong algorithm needs an interface that makes its decisions easy to understand.
What's next for LIFELINE
In the future, we want to add footage from park rangers' body cameras. Using computer vision, we'll scan that footage for signs that someone passed through, such as broken branches, flattened brush, and other disturbances in the foliage. Each detection becomes new evidence in our model, so the probability map sharpens as rangers cover ground and the drones automatically refocus on the areas that matter most. Combining eyes in the air with clues from the ground should find lost hikers even faster. Beyond that, we hope to test LIFELINE with real drone hardware and to work with search-and-rescue teams to tune our model against real cases.
Built With
- express.js
- javascript
- mongodb
- numpy
- python
- react
- recharts
- tailwind
- vite
- websockets
- zeromq