top of page
Ashwat Prasanna
PyroGuard
Rethinking firefighting with aerial precision and integelligence!
Inspiration
It all started with a fire: a devastating fire that reduced nearly an entire city to rubble and ash. In cities in India, like Bangalore and Mumbai, high levels of congestion and poor road infrastructure makes many urban fires inaccessible. Moreover, remote areas of villages have large factories, farms, and warehouses, where fires may go unnoticed for hours. A solution would provide a novel yet low-cost way of approaching fires in inaccessible regions autonomously.
Deadly fire in firework factory (TN) cannot be put out since it can't be accessed behind stacked inventory
First Test run (with timed payload deploy) of PyroGuard
AI Model Design
In order to keep the payload of the drone low-cost, a linear input was used in the form of a thermistor to identify higher temperatures from smoke. This was then passed to a logistic regression model which dynamically groups data into groups of similar input (like a KNN model) to get a wider range of possible success rates before running the logistic regression. The result, after training, was consistently very accurate (but for safety reasons could not be tested in a full-scale fire).
Introducing PyroGuard
Over the last decade, the implementation of drones for large-scale agricultural solutions has allowed this platform to become much more accessible to rural areas in India. Retrofitting computer vision models that can identify fires (and even fire hazards) to these drones, along with lightweight payloads, can help identify and mitigate fire risks.
Hardware
The hardware for the product was intended to be low-cost. The overall cost of the retrofittable attachment here was approximately USD 15 (INR 1200). This included corrugated plastic bases, small water containers (to be filled with firefighting explosives), and an Arduino-actuated servo mechanism to dispense the payload. The rest is hot glue and old wires.
Next Steps
Firstly, this proof of concept can be scaled up with larger drones and payloads, which will allow for in-situ testing (e.g. agricultural drones). Moreover, the use of a cloud VLM can be considered (as many areas in warehouses and factories have stable internet connections, this is feasible), which will be able to also identify potential fire hazards and pre-emptively suppress them.
bottom of page