The edge case
Recently, Insider explored the much-hyped world of edge computing, a market that some analysts expect to hit $250 billion by 2024. For the uninitiated, edge computing refers to a computing model where processing is done “closer to where data is being created in the physical world” instead of using data centers for cloud processing. To put a finer point on it, “edge computing focuses on smartening up the car, robot, or other systems right on the device, or placing a processor in closer proximity.”
What it solves for spatial computing
Think of an autonomous robot or drone in a commercial setting, like a factory or a construction site. To navigate its environment, it needs to gather data from its sensor payload and process it using a simultaneous localization and mapping (SLAM) algorithm, and possibly a number of other computer vision algorithms for object recognition, navigation, and so on.
If the robot or drone relied on cloud computing for this application, it would run into a few huge problems. The first is latency, which would slow down the operation of the device due to physical limitations on data transfer speeds. Another is the cost of bandwidth, which can get huge when you decide to use multiple robots in your factory and end up sending huge amounts of data to a center far away from you. It could also face serious problems with persistent connectivity.
A buzzword and a necessity
As The Verge helpfully explained it a few years back, edge computing may be so hyped as to seem meaningless—but it may be absolutely necessary for the advancement of technology like self-driving cars.
“Due to latency, privacy, and bandwidth, you can’t feed all the numerous sensors of a self-driving car up to the cloud and wait for a response. Your trip can’t survive that kind of latency, and even if it could, the cellular network is too inconsistent to rely on it for this kind of work.”
Of course, as with any new technology, edge computing will raise its own problems, such as lack of control over devices, regulation, update management, and so on.