Duke robotics research team creates chip to expand robots' motions
Source: Ajay Desai
Daniel Sorin, W.H. Garnder Jr. professor of electrical and computer engineering, helped design a new prototype chip that allows robots to make more efficient motion decisions.
A team in Duke's electrical and computer engineering department has made strides in robotics research by developing a prototype chip capable of helping robots plan efficient motions.
The effort was led by George Konidaris, assistant professor of computer science and electrical and computer engineering and director of the Intelligent Robot Lab, and Daniel Sorin, W.H. Garnder Jr. professor of electrical and computer engineering. Ph.D. student Sean Murray and seniors Will-Floyd Jones and Ying Qi have also been involved in the project.
Their work thus far has culminated in a new chip, which the team said can perform large sums of computations in order to choose a robot's motion path, solving a very unique problem in the field of robotics. Sorin explained that until now robots have typically been confined to set motion paths or have been burdened by time-consuming motion computations. However, the chip developed by Duke researchers is efficiently designed to allow robots to make quick and intelligent motion decisions, a skill that humans take for granted.
"For robots [motion planning] is really hard, and it turns out that this is a very computationally intensive problem," Sorin said. "Effectively, what is happening is that the robot is considering all of the possible motions it could do and trying to figure out [if each motion could] cause [the robot] to collide with an obstacle in its environment. Doing all of those tests to see if a motion will cause a collision requires a vast number of computations on a typical computer.”
Previous attempts at improving how robots move have used high performance general computers, but these solutions have been very slow due to the number of computations needed in order for a robot to work out a motion plan, Sorin noted.
Some have tried to accelerate the motion planning process with high-powered processors used for gaming, but Sorin explained that these processors consume a lot of power on a level that may not be viable for many vehicles or manufacturing companies, which might employ automated robots on an industrial scale.
“High power chips that consume 200-300 watts of power can be very prohibitive in many situations such as a [drone] or a factory," he said.
Sorin explained that these many limitations contributed to his team's inspiration for the new chip prototype.
“We thought that if we weren’t able to use an off-the-shelf part, could we build special purpose hardware to solve this problem? And that is what we did,” Sorin said. “It is so special purpose that it does nothing else but motion planning. But by virtue of only doing that one thing, it can do it really, really well.”
According to an article on the Pratt School of Engineering's website, the chip runs a complex algorithm that can check thousands of potential movements simultaneously. It has been successfully tested on a robotic arm, and the team has started to consider future applications for the chip, including autonomous cars and autonomous unmanned aerial vehicles.
One of the next steps for the team will be to make the chip viable for industrial use. The group is currently working on transitioning the chip from a general prototype used only in laboratory testing to a more specialized chip that can be modified for commercial use.
On the heels of previous projects from the Sorin group that have successfully made their way into the commercial world, their new chip is already beginning to receive attention from the commercial community. At the 2016 Robotics: Science and Systems Conference at the University of Michigan, Ph.D. student Sean Murray presented the group’s research paper and said their work caught the eye of several industrial companies.
“We got a lot of interest from Honda, Dyson, DoraBot and Mitsubishi. It’s still in the early stages, but we made a lot of connections at [the conference] to help with that process," Murray said.
The development of the chip is related to the group's startup company called Realtime Robotics. Sorin noted that their entrance into the field of commercial robotics has led to preliminary discussions with other robotics companies in rolling out their research on an industrial scale, while the team continues to work on enhancing the chip.
| }
|