![]() |
(Image source: Microsoft) |
The next wave of automotive innovation depends on artificial intelligence (AI).
At his DesignCon 2017 keynote, Doug Seven, Microsoft's Group Program Manager – Things That Move, told the audience that in his meetings with automotive execs cars aren't being talked about like machines anymore. “Cars are essentially data centers on wheels,” Seven said.
Seven said there's enough pressure on the automotive industry right now to make changes toward autonomous cars, but there are also key innovations that need to happen. Paramount among these for Microsoft is that cars need to become intelligent.
“When you think about what it takes from a software and hardware standpoint to make a car autonomous it's daunting,” Seven said. “It's not just about sensors and connectivity, it's about intelligence and low latency.”
Machine learning is already being applied to vehicles and holds promise in areas such as predictive maintenance and real-time analytics. At CES 2017, Microsoft unveiled the Microsoft Connected Vehicle Platform , a cloud-based platform that serves as a reference for automakers and design engineers to build solutions in five key areas: telematics and predictive maintenance; productivity and digital life; connected advanced driver assistance systems (ADAS); advanced navigation; and customer insight and engagement.
In addition to this, Microsoft has already released its Microsoft Cognitive Toolkit , an open-source toolkit for creating applications capable of deep learning across clustered environments, as well as the Microsoft Cognitive Services API.
Seven talked about how Microsoft's Cognitive Services API already includes an Emotion API capable of understanding the emotions of someone's face, including anger, contempt, disgust, fear, happiness, neutral, sadness, and surprise. He asked the audience to imagine applying this same capability to autonomous vehicles. “If we could detect things like road rage or stress, we could [have a vehicle] do things to alter the environment for the driver or passengers.”
![]() |
But that term “cloud-based” is the obstacle. Right now the Cognitive API and Cognitive Toolkit are cloud-based, which won't suffice for self-driving cars.
“What we can do in the cloud with our AI capabilities is to build algorithms and models to let cars become intelligent and make decisions,” Seven told the DesignCon audience. “But we can't rely on the cloud because we might lose connectivity or their might be latency issues."
The next step is to create AI within the cars themselves that will have meaningful interactions with drivers, passengers, and the car's environment. "We're talking about putting hardware in a car that is capable of processing data for the purposes of AI," Seven said.
“We have to move the workload from the cloud to the edge,” Seven said. “We need hardware in the car capable of processing that data in real time.” According to Seven, self-driving cars need latencies in the sub-millisecond range to be fully effective on
Comments
Given the track record
We can all hope that it's
Yes, given their track record
Bad idea. First of all,
The harsh environment for the
The really dangerous part
Add new comment