Wayanad landslides
400+ deaths in the Western Ghats. Roads cut, villages buried. Rescuers could not locate survivors under debris โ no hyper-local alert or through-wall sensing.
Wayanad landslide watch โ Kerala, Jul 2024. 400+ affected. No hyper-local mesh.
โฑ M4.2 seismic
Vellore region ยท awaiting approve
13:00 โ 13:451 person detected ยท motion YES
92% confidence ยท 4 nodes
Open-Meteo ยท LAN ยท WiFi CSI
Early warning for Indian communities, then offline WiFi CSI survivor sensing when networks fail.
South Asia faces earthquakes, landslides, and floods every year โ yet when disaster strikes, responders still cannot answer three basic questions: where are the survivors, how many are there, and are they still alive?
AURA closes this gap with affordable ESP32 sensing: early alerts before disaster, then WiFi CSI survivor detection after networks fail.
400+ deaths in the Western Ghats. Roads cut, villages buried. Rescuers could not locate survivors under debris โ no hyper-local alert or through-wall sensing.
Teesta River flash flood killed 40+. Remote valleys had no last-mile warning โ families did not know where to go or who was still trapped upstream.
600+ buildings cracked. Slow-motion disaster needed continuous monitoring of how many residents remained inside unsafe structures.
Cyclone Michaung submerged streets. No ward-level safe-house map โ people did not know which shelters had space or how many neighbours were stranded.
Magnitude 6.4 struck western Nepal โ 150+ killed. Remote hills had no village-level warning and no device to find people under collapsed stone homes.
Disaster photos via Wikimedia Commons (CC BY-SA / public domain).
One alert node monitors hazards. Four rescue receivers + one transmitter sense survivors without internet.
USGS + Open-Meteo โ DM approves โ LAN + webhook broadcast with safe-house map.
Motion, count, position, vitals on laptop via AURA_HUB hotspot โ no cloud needed.
Simulation lab, alert feed, DM console โ demo-ready for judges in under 60 seconds.
AURA combines early warning (before disaster) with rescue sensing (after networks fail). Here is how each part works.
The laptop (or one ESP32 alert node) watches your configured coordinates for earthquakes and severe weather.
Nothing goes public until a disaster manager approves it. You can change watch location on the Alerts or DM Console page.
When cell towers fail, ESP32 nodes form a local mesh. WiFi probe packets bounce off human bodies; receivers measure CSI to find who, where, and vitals.
Outputs: person count, XY position map, motion, respiration & heartbeat โ no internet required. Test with bundled act_105_48 in the simulation lab.
Channel State Information captures tiny changes in WiFi signals when a chest moves (breathing) or a limb shifts (motion). AURA fuses four receiver angles to estimate position inside a 10ร10 m zone.
USGS global earthquake feed + Open-Meteo wind/precipitation for your watch coordinates. Thresholds in disaster_alert/config.yaml filter noise; only significant hazards enter the DM queue.
Simulation lab replays real WiMANS CSI + video. DM Console simulates an earthquake alert and broadcasts to the public feed.
No hardware required โ run the simulation lab and DM console on this laptop.