Kenneth Richard Geipel, Co-Founder & CEO of Robotto, delves into FireAI’s use of edge-level AI for enhancing firefighting operations through autonomous drones and immediate situational update
I come from a background in the Danish Armed Forces, where I spent about ten years and had three deployments in Afghanistan with reconnaissance.
After my third deployment, I needed a change of pace.
So, I decided to pursue my other passion: robotics.
At university, I worked in a group with the people who are now my co-founders.
Robotto started as a spinoff from our bachelor thesis, where we developed a novel method to use neural networks on drones to calculate objects in 3D using a single camera.
Although I hold the title of CEO, I see myself more as a co-founder.
Our first product was in the wildfire space.
We realised that firefighters struggle with outdated data during wildfire operations.
Typically, the data they use is 12 to 24 hours old, often from satellites.
This delay means that every time they respond to a fire, the operation is delayed because they first need to gain an updated overview of the situation.
Development
The inspiration came from our work with edge-level AI on drones.
This means all data processing occurs locally on the drone, allowing real-time data processing and enabling the drone to act on the situation immediately.
We equipped the drones with small GPUs or processing modules, and we have a custom setup that communicates with the drone.
Here’s how it works: you’re given a map of the current area, you select the operation area, press go, and the drone operates completely autonomously from there.
It calculates the route to the area and within it, our AI looks for fire, smoke, or heat signatures through a thermal camera.
When the drone detects a fire, it maps it out in real time, providing firefighters with immediate situational awareness.
Real-time awareness
FireAI enhances traditional systems by providing real-time situational awareness.
Currently, firefighters struggle with outdated data.
When using drones powered by FireAI, first responders can deploy the drone to instantly map out the fire in the field.
Additionally, because our system works autonomously, drones can cover high-risk areas and identify fires as they occur, alerting local security.
Our software has the potential to integrate with larger platforms, such as fixed-wing aircraft and other drone platforms, to monitor extensive areas continuously.
This setup ensures firefighters receive immediate data whenever a fire is detected.
Edge-level processing
Our differentiation lies in our use of edge-level processing.
The entire system is designed to be completely autonomous, with no data transferred back and forth.
All our navigation is visually based, meaning the system can be deployed anywhere in the world, even without satellite or other connections.
You simply provide an operational area for the drone, send it out, and it handles everything.
The drone identifies fires, calculates their locations, and measures wind speed and direction.
This allows the drone to navigate safely around fires, mapping them out before surveying the rest of the area.
We mount a GPU on the drone, sending all telemetry data and camera feeds to it.
This enables the drone to understand its situation and respond accordingly.
If it would help, I can send you a video from some live operations in Spain, so you can see how the end user experiences it.
A GPU, or Graphics Processing Unit, is a specialised processor designed to accelerate graphics rendering.
In the context of FireAI, the GPU is used to handle complex calculations and data processing on the drone itself.
This allows the drone to perform tasks such as identifying fires, calculating their locations, and navigating autonomously in real time.
By using a GPU, the system can operate efficiently and independently without needing constant data transfers to and from a central server.
Post-operational review
While the news often focuses on large wildfires with huge flames and extensive fire fronts, a significant part of wildfire operations happens after these major events.
This is known as the mop-up phase.
During this phase, firefighters need to search the entire area for hotspots that could reignite.
Traditionally, this involves combing through the area with their resources to check for pockets of thermal output, or hotspots.
Using our technology, we can detect hotspots down to a diameter of 15 centimetres and map them out.
This allows firefighters to survey these areas much faster using an aerial view.
They can then focus their efforts on specific locations, putting a hose in the ground to extinguish any remaining fires.
FireAI has a KML export function.
This essentially creates a map of the fire, generated from the drone’s data.
Firefighters need to share this fire map with their team, so we integrated the function with various services, including WhatsApp.
This allows firefighters to send the KML files directly from the drone controller to anyone who needs it, enabling instant sharing within the organisation.
During hands-on trials with our collaborators at Bombers GRAF, we prioritised perimeter mapping.
This was in response to the pressing challenges of vast fires and prolonged delays for updated fire maps.
However, as emergencies evolved, the importance of post-operation assessments surged.
We introduced a feature targeting the detection of smaller hot spots, which might pose risks when exiting a site.
Unlike the perimeter mapping that contours the fire’s boundary, hot spot detection works on the premise that there aren’t significant heat sources.
It focuses on flying above and pinpointing these hot spots, integrating them into a map for detailed analysis.
User response
We co-developed FireAI with the Danish Emergency Management Agency and several fire services, including NASUWT and cattle and firefighters.
We discovered that the firefighting industry is quite traditional, often still using methods from the 1980s.
One of the main challenges we face is the adoption of this technology, especially because it involves autonomous drones.
There is some hesitancy towards using autonomous systems due to risk assessments.
For instance, concerns arise about potential conflicts if an air tanker or another aircraft enters the same airspace.
To address this, we designed the system so that users can take manual control of the drone whenever necessary.
However, we have also encountered resistance from manned aircraft operators who perform similar fire mapping missions, showing some pushback against this new technology.
Future firefighting
Our team is currently in Torino working on the next iteration of FireAI.
Since wildfires are seasonal, we want to create a tool that firefighters can use year-round.
We are expanding its capabilities to include search and rescue operations, where it can detect personnel in the wild.
Additionally, we are incorporating flood mapping into the system.
Moreover, we are currently integrating FireAI into fixed-wing aircraft, extending the flight time from about 30 minutes to three hours.
This is a significant improvement, and we hope it will provide enough value for firefighters to adopt the technology more widely.
About the author
Kenneth Richard Geipel is the Co-Founder & CEO of Robotto, a multi-award-winning startup integrating drones with AI and computer vision technologies.
He started his career in the Danish Armed Forces, becoming the youngest Corporal and receiving the Armed Forces Appreciation for Extraordinary Service.
Kenneth pursued a career in robotics, leading to the creation of Robotto from a bachelor thesis in 2018.
Robotto focuses on developing autonomous, edge-computing systems for real-time data processing in various critical applications.
This article was originally published in the July 2024 issue of International Fire & Safety Journal. To read your FREE digital copy, click here.