Building smart cities by harnessing the power of edge computing

Simon Crosby has a fun fact about smart cities and information: “Every day, the city of Palo Alto streams more data than Twitter.”

He should know. As CTO of Swim.AI, a Silicon Valley-based artificial intelligence company founded in 2015 to help cities optimize sensor data, Crosby worked directly with Palo Alto — a leader in embracing smart city technologies that has been recognized by the US Center for Digital Government as a Top 10 Digital City — to improve its traffic infrastructure. It’s just one example of how cities are increasingly harnessing the power of AI to run more efficiently.

There’s good reason for cities to want to become “smart”: Research by McKinsey Global Institute suggests they can reduce commute times by up to 20% and even reduce fatalities by up to 10%. As they connect everything from traffic signals to electrical grids, sensors in this growing Internet of Things (IoT) are already pushing out enormous volumes of data.

Meanwhile, industrial IoT devices are generating information about equipment such as wind turbines, industrial robots and magnetic resonance imaging scanners.

There’s enormous potential in using the data to enhance efficiency and safety, but making the best use of the data requires edge computing and intelligence — the ability to aggregate and analyze data close to where it’s captured.

Making sense of IoT

“[Microprocessor supplier] ARM ships 5 billion processor cores, which allow a computer’s CPU to process multiple tasks at once, per quarter — you take them and all the other device vendors, and that’s a lot of devices,” Crosby remarked in a speech at the 2018 Rakuten Technology Conference, referring to chips that power IoT devices. “How are these edge devices going to get smart? By sending their data up to the cloud? Amazon, Google and Microsoft would love that. I’m here to tell you there’s another way: by running around, making experiments and learning on the fly.”

Crosby told attendees about the increasing importance of incremental learning, an approach to machine learning that uses input data to better train a software model. The result is the “intelligent edge,” Crosby said, that “analyzes and learns and predicts from time-series data an evolving worldview.”

Swim.AI, which closed a $10 million Series B funding round backed by ARM last July, builds and self-trains “digital twins” of edge devices that can analyze, learn and predict in real-time, as well as share insights with other twins in the network. A real-world example from Palo Alto, California, would be using traffic signal sensor data to predict traffic flows, and by using only minimal, low-cost hardware.

“With a $200 NVIDIA Jetson device, which has a 256-core GPU, I can do something totally awesome,” Crosby said. “I can route a Waymo [self-driving] car through the city and it will never stop. That is, the digital personas of each one of the intersections can predict far enough out what the traffic’s going to do, so I know exactly when the phase of each light will be green.”

Simon Crosby, CTO of Swim.AI, on stage at Rakuten Technology Conference in Tokyo in October 2018.
Simon Crosby, CTO of Swim.AI, on stage at Rakuten Technology Conference in Tokyo in October 2018.

Building edge intelligence on existing legacy systems

Earlier this year, Swim.AI and Texas-based traffic control system company Trafficware launched TidalWave, a live streaming traffic data service powered by edge computing. Rolled out nationally, it features “precise, granular traffic data at a resolution of hundreds of milliseconds, at a small fraction of the cost of central cloud-hosted learning and prediction,” Swim.AI CEO Rusty Cumpston said in a release.

Smart cities and traffic prediction are an obvious application. But Swim.AI has also brought its Swim EDX technology to manufacturing, using edge intelligence to track the many parts and RFID sensors used in building military aircraft, and to predict errors in the assembly of computer systems. Another application is optimizing the distribution of electricity for utilities to best meet demand.

“The encouraging news is that for a large number of legacy systems — assembly line robots, the works — you can learn on the fly and predict things that are of great interest, either to applications or to humans,” Crosby shared in an interview on the sidelines of the conference. “With many of these systems, the value of the data is ephemeral and it rapidly decreases over time… The key point is to derive value from data as quickly as you can and then throw it away.”

With his insights into the growth of smart cities and smart manufacturing, Crosby, also a cofounder of California cybersecurity firm Bromium, offered some predictions about how our world will change.

“I think what we’ll see is that the world will get smarter all around us, but there isn’t this all-seeing, all-dominant intelligence,” Crosby added. “That’s a very hard problem to crack. So, things will get better fast. My key point is to say, ‘Yes, it can be done, and on existing hardware, with just a bit of software.’”


For more about Rakuten Technology Conference 2018, visit here.

Tags
Show More
Back to top button