February 17, 2025
This article continues my earlier exploration of Agentic AI. If you missed that piece, you can check it out here: https://www.linkedin.com/pulse/agentic-ai-ooda-loop-how-could-support-fire-service-jonathan-boyd-gjihc/?trackingId=E7w9jbrbCUILYiqKz3laug%3D%3D
Time remains an unyielding adversary. In our operations, every fleeting second counts. Shifting mitigating actions closer to the incident’s onset has the potential to change outcomes drastically. Agentic AI holds promise in this regard; yet its efficacy depends on a supporting partner, edge computing. When an incident occurs, waiting for data to traverse to a distant server and back is simply not feasible. Emergencies come with their own array of challenges: damaged infrastructure, intermittent connectivity, and restricted bandwidth compel us to process data at its origin.
Read about DHS testing of Edge Computing for wildfire: https://www.dhs.gov/science-and-technology/news/2024/08/08/feature-article-edge-computing-wildfires-edge#:~:text=UC%20San%20Diego%E2%80%99s%20WIFIRE%20Lab,data%20collection%20and%20processing%20tools
What is Edge Computing?
Edge computing refers to the practice of processing data on or near the device that collects it instead of sending it to a remote server. Picture a small computer right at the scene, analyzing inputs without delay. A drone, for instance, might filter its own video feed and only dispatch an alert if it registers an anomaly, rather than transmitting every frame. This approach converts streams of raw data into immediate signals and preserves bandwidth.
Agentic AI operates by observing, orienting, deciding, and acting in a self-reliant cycle reminiscent of the OODA loop. For such a system to perform its duties, it must access information quickly. Edge computing supplies that immediate access; data is processed where it is generated, enabling an AI agent to issue warnings and make decisions right then and there. Even if external communication fails, the local processor continues its work, ensuring that no critical update is lost.
Hardware progress in this area is noteworthy. NVIDIA offers compact platforms like the Jetson series, which deliver significant computing capability in a small package suitable for integration into drones, cameras, or wearable devices.
Read more about NVIDIA Jetson: https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/
NVIDIA® Jetson Nano™ makes it possible to bring incredible new capabilities to millions of small, power-efficient AI systems. It opens new worlds of embedded IoT applications, including entry-level Network Video Recorders (NVRs), home robots, and intelligent gateways with full analytics capabilities.
Other manufacturers are crafting specialized chips for on-site analysis, while frameworks such as TensorFlow Lite and PyTorch Mobile allow AI models to be deployed on devices carried into the field. Such innovations have begun to shape local processing capability.
Early products illustrate what may come. The Qwake Technologies C-THRU system embedded in firefighter helmets processes thermal imaging and sensor inputs on board, providing immediate visual cues in smoke-obscured conditions.
Tethered drone systems, equipped with on-board analysis, can scan for hotspots and alert command when necessary.
Additionally, systems like the 3AM Innovations Inc. Florian system are being crafted to manage the overload of incident command data. In a scene filled with blaring radios, overlapping voices, and incessant phone calls, 3am Florian transcribes voice communications and flags any MAYDAY calls, notifying the incident commander.
Applications of edge AI could extend beyond firefighting and into EMS. Lifepak EKG monitors and portable ultrasound systems could process data locally, offering immediate diagnostics for patients. Drone-based damage assessment tools could analyze a disaster zone on site and report status updates quickly. There is promise in SCBA telemetry systems, where a local AI agent could track air levels and issue warnings when safe limits are breached.
Yet challenges persist. Many current AI systems still lean on remote processing; if connectivity falters, performance may suffer. The reliability of autonomous agents is not absolute, errors can occur, and scaling them to handle full-scale emergencies remains a work in progress. Security issues arise as field devices are exposed to both physical and cyber risks. Moreover, the flood of data from numerous sensors demands a method to isolate vital alerts from background noise. Local processing, by filtering data before transmission, offers one path forward.
Until local processing reaches complete maturity, portable cloud solutions are emerging as an interim measure. Devices such Azure Stack Edge could act as mobile mini-servers positioned close to the incident. These units handle intensive processing when links to central servers are weak, bridging the gap between full cloud reliance and on-site data handling.
I think it would be smart to start preparing for these systems today. When acquiring new hardware, tablets, radios, mobile devices, ask about their computing capacity, GPU specifications, and memory limits. Invest in equipment capable of supporting future edge AI applications. Engaging in pilot projects and early trials can shed light on how these systems integrate into our operations. Collaboration with technology providers and research teams will help shape tools that meet our specific needs in the field.
Edge computing and Agentic AI represent a shift in how emergencies will be managed. Time is our enemy. Every second counts. With edge computing and Agentic AI processing data on site, we save crucial moments. Sensors detect changes and local AI responds immediately without waiting for remote servers. This quick response gives incident commanders the information they need at the right moment. As we integrate these systems into our operations, we move our decision-making closer to the start of an incident, putting time on our side and improving outcomes.