Why Wireless Sensing Is Edge AI’s Next Act
Rajesh Subramaniam is Founder and CEO of embedUR systems.
gettyEdge AI has become nearly synonymous with computer vision, with cameras detecting people, objects and faces on-device. But vision is only the first sensory modality to reach the edge, and in many settings, it’s the wrong one. Cameras don’t work as well in darkness and can exceed the power budget of battery-operated devices.
Wireless sensing offers another, more powerful approach. Using radio signals and devices already in place, it can detect movement, human presence and changes in the environment without requiring any additional sensors. This capability could turn billions of existing devices into ambient sensors, opening up new possibilities for edge AI.
The radio chips in routers, hubs and light switches already communicate over Wi-Fi, Bluetooth, cellular, UWB and Zigbee networks. The same wireless signals used for communication can also be used to sense what’s happening in the surrounding environment, with no camera and no wearable required. When a body moves through a room, it disturbs the signal paths between those devices, and software interprets the change in pattern. This makes it possible to sense human presence and activity without installing any additional infrastructure. Today, wireless sensing is mostly used for basic motion and presence detection, with commercial examples including Comcast WiFi Motion and Philips Hue MotionAware.
What’s new is that edge AI models running on low-power microcontrollers can now interpret these signals in real time. These models are trained on a substantial amount of data showing how signal patterns change when somebody walks or when a signal hits furniture or a wall and bounces back. Edge AI adds significant value because compute and inference run locally, with no cloud round-trip and low latency. The system also keeps working during internet outages.
This unlocks applications in uses ranging from smart buildings to elder care. Real-time occupancy detection can help optimize heating, cooling and lighting usage, with significant potential energy savings. (Motion-triggered lights are already the leading use case.) Building managers can also understand when and how offices, meeting rooms and parking spaces are used. In the security space, wireless sensing can detect unexpected after-hours activity and support asset monitoring, such as whether containers or storage units with valuable contents are open or closed.
More advanced capabilities include fall detection for elder care and breathing rate tracking even when someone is standing still. These exist in demos, research and niche care products using high-precision millimeter-wave radar chips, but they aren’t yet a mass-market capability on ordinary routers.
Developers are focusing on accuracy, with a lot of validation and testing, so operators can trust the data and results. We definitely don’t want false occupancy readings from things like pets, robot vacuums and oscillating appliances undermining trust in wireless sensing systems. Coverage and interference are a concern, too, since performance can vary widely by environment. Distance and walls can weaken signals and add multipath interference.
Privacy is also extremely important, since tenants, employees and consumers may object to constant monitoring. One mitigation could be to sense only when necessary rather than continuously. On the other hand, edge-only inference limits data exposure since raw sensing signals can remain on the devices. Reducing exchange between the edge and the cloud cuts the odds of private data such as people’s activity patterns or medical information falling into the wrong hands.
I fully expect this technology to be mainstream and deployed in millions of devices over the next five years, resulting in huge operational efficiencies and cost savings. Every type of common consumer device will eventually have some sort of edge processor, including smoke alarms, carbon monoxide monitors, lights and fans. Passive infrared motion sensors, now ubiquitous, will become obsolete, so there will be fewer devices to buy, install and maintain. The smartest rooms of the future may be the ones with no cameras at all.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
