Machine Learning at the Edge

August 4, 2023 | By Ari Basen

machine learning

The proliferation of internet-connected devices and sensors in the Internet of Things (IoT) era has created an explosion of data. However, transmitting all this raw data to distant cloud data centers for processing introduces latency, bandwidth bottlenecks, and reliance on connectivity. For time-critical applications like autonomous cars or remote sensor grids, cloud-centric architectures are insufficient. If an orbiting satellite must wait for a cloud-based AI model to interpret potential collision data – it would be halfway around the Earth by the time it got an answer, striking everything in its trajectory in those critical seconds.

What is Edge Computing?

Edge computing enables immediate analytics processing data right at the source – whether on an oil rig, the factory floor, a vehicle, or on a farm. By analyzing data at the source, edge computing provides ultra-low latency, reduces bandwidth needs, and ensures continued operation even if connectivity to the central cloud is lost. Rather than relying on a remote cloud, the intelligence, and responsibility are distributed across the entire network. Edge computing complements cloud computing with dispersed, localized computing power for when milliseconds matter, bandwidth is scarce, or systems must function independently.

New Possibilities with Edge Computing

Programmable Logic Controllers (PLCs) have provided basic computational abilities and automation to machines locally for decades. Though widely used – from factory floors to bridges to streetlights – PLCs have limited capabilities. They execute predefined logic for basic tasks like turning equipment on/off or setting speeds. PLCs lack awareness beyond their narrow functions, are hardcoded in proprietary languages, and have little ability to communicate or interoperate with other systems.

In contrast, edge computing leverages more versatile internet-connected devices. Edge nodes can run software for dynamic data analysis and machine learning, not just control logic. Critically, edge devices can talk with each other and exchange data using standard protocols like HTTPS, even across vendors. This enables coordinated analytics and automation between distributed assets. For example, sensors along a pipeline can collectively detect and mitigate dangerous pressure levels by adjusting valves, without relying on a central system.

Edge networks are also resilient. They can continue functioning locally even if disconnected from centralized infrastructure, by pooling data across edge nodes. This makes edge computing ideal for environments with limited connectivity like oceans or space. In essence, distributing intelligence to the network edge enables faster, more coordinated, and more autonomous responses to situations without constant human intervention. This is a paradigm shift from centralized client-server architectures and into an intelligent distributed networks that can improvise and self-adapt.

Current Limitations of Edge Computing

The relative novelty of edge computing has led to fragmentation in hardware and communication standards, hampering interoperability. In 2023, over 10 different protocols are used in IoT and edge devices, including Wi-Fi HaLow, Bluetooth LE, ZigBee, LoRaWAN, MQTT, and more. These operate on different radio spectra, preventing cross-compatibility even if packet formats are standardized. As the ecosystem matures, consolidation around dominant standards like Wi-Fi and Bluetooth is likely.

The lack of dominant standards also hampers deploying and managing edge devices. In cloud environments, Docker and Kubernetes emerged as standard tools for containerized apps and orchestration. But for edge ecosystems, few mature management platforms exist despite a plethora of options. Kubernetes recently developed KubeEdge to manage edge nodes by modeling devices as digital twins. However, KubeEdge is tailored for centralized management from a single computer, not decentralized coordination.

AWS IoT Greengrass is another offering, allowing serverless Lambda functions and Machine Learning (ML) models to be run locally on devices. Its advantage is minimizing cloud communication for security, latency, and cost. But the lack of convergence around solutions like Greengrass inhibits portability across fragmented edge stacks. Until dominant orchestration tools emerge, complexity and fragmentation will hinder unlocking the full potential of edge computing.

Machine Learning and Edge Computing 

A key advantage of Edge devices beyond the traditional PLCs is that they can learn, from themselves and one another. One common application is the Nest Learning thermostat found in many homes. It not only connects to the user’s phone to be set remotely, but it also takes in weather data, temperature sensor data, and positioning data from the user’s phone to learn the routine of the household. It knows when you are home and when you’re at work, and can predict when to preheat or precool your home automatically so it’s room temperature right when you get home. Taken together as a larger network, these learning thermostats can be used to predict citywide demand for electricity at an industrial scale when combined with smart meters. These Edge ML devices not only learn local patterns but can feed a larger ecosystem to help predict major spikes in electrical demand and help automatically load balance to help prevent blackouts. These “Smart grids”, like those deployed in Austin TX, harness edge ML to forecast electricity demand across regions, detect failures, and optimize power distribution. The grids self-adjust based on emerging supply/demand imbalances. The metro system in Hong Kong utilizes thousands of sensors deployed throughout the system to predict breakdowns and traffic bottlenecks. By utilizing Edge ML to monitor and predict breakdowns, the Hong Kong metro AI can then proactively schedule and synchronize a weekly 2600 maintenance work-orders and maintain an astonishing 99.9% on-time record. 

National Security Implications and Opportunities

The same predictive analytics which allows carmakers like Ford or Tesla to upgrade their vehicles over the air (OTA) can also be applied to military vehicles like an Abrams tank or Apache gunship. As Ford CEO Jim Farley explained, the mentality used to be “We’ll sell you a car and we will see you in four years” and now it is sifting to one of the constant over-the-air updates and making the vehicle “fully connected” like a smartphone. Such OTA updates work well in a major metropolis but would fail in the electromagnetically challenged environment of the modern battlefield. 

Future weapons platforms will need embedded Edge ML models to provide situational awareness and autonomy. For example, an armored vehicle could use onboard sensors and Edge ML to detect signs of potential failures like overheating barrels or damaged treads. The vehicle can then compensate locally by adjusting the firing rate or movement. Additionally, by processing data at the edge, issues can be identified early and preemptively repaired at forward operating bases, rather than waiting for major centralized software updates from defense contractors.

These edge learning systems also enable fleet-wide improvements in near real-time. The data from edge models across units are aggregated in a centralized cloud analytics platform. Here, reinforcement learning algorithms continuously optimize designs by analyzing battlefield failures and challenges experienced across the fleet. This allows smarter iterations of parts like engines, armor, and weapons based on emerging needs, far faster than traditional delayed development cycles.

In essence, pairing edge ML for localized awareness and adaptation with fleet-wide cloud analytics enables both individual systems and the platforms overall to rapidly evolve and improve while in the field. This accelerates the development cycle from years to days.

Decentralized Autonomous Organizations (DAOs) for the Intelligent Battlefield

Decentralized Autonomous Organizations (DAOs) represent the cutting edge of networked Edge ML. DAOs move beyond basic swarm intelligence, where groups of drones fly in formation, to enable deeper multi-system coordination. DAOs allow heterogeneous devices like sensors, vehicles, and satellites to establish shared situational awareness across domains. This facilitates complex joint operations involving diverse assets. Rather than top-down control, DAOs distribute intelligence to unlock new possibilities for autonomous, resilient missions.

In the current war in Ukraine, a DAO would connect seismic sensors trained on the unique vibration fingerprints of Russian tanks and then automatically cross-cue this with overhead UAVs that use computer vision to gain positive ID on the target. This information is then synchronized with an artillery battery which uses its own Edge ML to compute a firing solution giving expected collateral damage, windage, distance, and known characteristics of the targets. The C2 synchronization of these various battlefield assets rapidly accelerates the sensor-to-shooter cycle. Human commanders would oversee these operations, but orders would no longer need to be given to each asset and then processed at the command center. Instead, a DAO powered by various Edge ML models could orchestrate such combined arms maneuvers on its own.

A DAO for naval operations in the South China Sea could coordinate ships, manned aircraft, autonomous undersea vehicles (AUVs), and sensors to hunt enemy submarines. Passive sonobuoys dropped from P-8s or hydrophones installed on the seafloor detect submarine acoustic signatures identified using onboard ML models so they are not transmitting data and thus can remain undetected. The hydrophone network leverages ML to predict the submarine’s path, and cues adjacent hydrophones to activate and monitor along the projected route. Nearby surface vessels dispatch underwater drones that use edge ML on hydrophone streams to classify the contact and localize it precisely. This data fuses into a common operating picture to guide a coordinated attack by the surface vessels and aircraft with the DAO optimizing actions of all assets. The DAO’s distributed nature provides resilience if any node gets disrupted.

High above the Earth, a space defense DAO could network diverse assets for real-time missile tracking and interception. Infrared early warning satellites detect launches and instantly classify missiles and predict trajectories using edge ML, rather than relying on delayed ground analysis. The DAO would correlate these alerts to cue electro-optical sensors and airborne platforms to maintain eyes on the target. By fusing sensor data from space, air, and ground, the DAO maintains persistent surveillance of missiles in flight to continuously update impact predictions and recommend optimal actions, like launching interceptors or repositioning disposable satellites to block. Rather than isolated systems, the DAO seamlessly orchestrates ground, air, and space-based ISR to provide a unified view of the evolving situation. The space defense DAO allows for rapid, automated, coordinated responses for targets moving at orbital velocities that would challenge the reaction time of even the best team of human defenders.

How Oteemo Can Help You Harness the Power of Edge ML

As edge computing and distributed AI transform national security operations, partners like Oteemo are needed to deliver and integrate these robust edge capabilities. Oteemo specializes in rapidly prototyping and deploying Edge ML systems tailored to military requirements. Our experienced engineers and developers design customized Edge ML solutions for defense applications requiring localized real-time analytics and resilience. We fuse domain expertise with technical excellence to solve complex challenges at the tactical edge. Whether optimizing sensor fusion, enabling predictive maintenance, or coordinating through DAO architectures, Oteemo is the partner needed to bring defense edge capabilities from concept to reality. 

Contact us today to discuss how distributed learning at the edge can give America’s warfighters a decisive advantage. We look forward to working hand-in-hand to usher in a new generation of intelligent systems and deliver unmatched edge computing capabilities for 21st-century defense.


Submit a Comment


Who We Are & What We Do

As passionate technologists, we love to push the envelope. We act as strategists, practitioners and coaches to enable enterprises to adopt modern technology and accelerate innovation.

We help customers win by meeting their business objectives efficiently and effectively.

icon         icon        icon

Newsletter Signup:

Join tens of thousands of your peers and sign-up for our best technology content curated by our experts. We never share or sell your email address!