In the News
SAN JOSE, Calif. — Thanks to cloud robotics, both autonomous vehicles and robots should be able to learn from one another and become more intelligent collectively, said James Kuffner, chief technology officer at the Toyota Research Institute.
“We’re developing technology for automotive safety and robotics,” Kuffner said. “We’re also looking at helping the aging societies of Japan and elsewhere.”
He described progress in artificial intelligence, sensors, manipulation, and cloud computing as following historical patterns and enabling safer, cheaper, and more reliable devices.
Cloud robotics to be more agile, autonomous
“In the next 50 years, we’ll see robot intelligence be useful beyond factory automation,” Kuffner said. “Autonomy is important in logistics, in the home, and for maintaining the quality of life for aging in place.”
“Search-based AI,” such as Google’s AlphaGo and route-planning algorithms for robots, are important steps toward autonomy, he said.
“Planning dynamic actions isn’t just for bipeds; it includes a compound motion such as a leap and jump for quadruped robots,” Kuffner said.
“The big question is, ‘When are we going to have an intelligent robot personal assistant?'” he asked. “Three things are holding that back: capability, cost — but sensor technology is rapidly progressing, as well as safety and reliability.”
Overcoming obstacles by moving to the cloud
“In teleoperation, humans provide a ‘remote brain’ for robots,” Kuffner explained. “We can now move that intelligence instead to the cloud.”
“Cloud robotics enables cheaper, lighter, and smarter robots,” he said. “My four kids have spilled drinks many times, and I wish I could have transferred the knowledge between their brains instead of waiting for each one to learn in turn.”
“Big data and machine learning will usher in a new era of advancing robot and vehicle autonomy,” Kuffner added. “Thanks to the cloud, small companies can access modern compute and data-storage resources.”
It’s not yet possible to include all the machine-learning capacity desired within a robot or an autonomous vehicle, but smartphones and the cloud “open possibilities,” he said.
“Smartphones turn over every 18 months, and they include everything you need for a robot — IMU [inertial motion unit], GPS, 3G, Wi-Fi antennas, camera on a chip, and storage,” Kuffner said. “With ‘Build Your Own Cellbot,’ we made it easy to program and hack robots.”
“Who would have thought that you could stream video to phones?” he said. “Moore’s Law pales in comparison to improvements in wireless broadband speed. That opens possibilities for cloud robotics.”
According to Kuffner, the potential benefits of cloud robotics include:
- A shared knowledge database, useful for object recognition and mapping
- Offloading heavy computing tests to the cloud and easier updates of software
- A skills/behavior database, allowing for gathering and data mining of collective experiences, and a library of actions to help all robots, such as in speech and language translation
“There’s no reason why robots can’t understand all languages,” Kuffner said, showing a slide of Star Wars‘ C-3PO.
Advances in image recognition, deep learning, and 3D models of objects can help with grasping and manipulation, he said.
“Human crowdsourcing can scale hard semantic and quality-control problems globally, so we can enable ‘robotsourcing’ with large-scale deployment of data-sharing robots,” Kuffner said.
“By sharing and maintaining data in real time, you can have an updated map of the road surface scalable to all vehicles,” Kuffner said.
One attendee noted that many people in Japan are skeptical of...