Helping Robots To Help Themselves
Goldberg's alternative is to give robots the ability to parts-feed themselves. In collaboration with two Silicon Valley robotics companies, and backed by a grant from a California state agency, he is developing a promising prototype, built in a USC lab, into a robust system for industrial use.
"When setting up an automated assembly line," Goldberg explained, "the robots can typically be programmed in a matter of hours. But creating a parts feeder to get a properly positioned part into a robot's grasp is a 'black art.' It can easily take weeks or even months to get one running properly."
In collaboration with Brian Carlisle, chairman and CEO of Adept Technology Inc., of San Jose, Goldberg and his students at the School of Engineering's Institute for Robotics and Intelligent Systems (IRIS) have developed a new approach.
"We are creating a system in which parts can just be randomly dropped on a conveyor belt," said Goldberg, associate director of IRIS. "A video camera eye looks at each part as it comes along. That video image is rapidly analyzed to direct a gripper to pick the part up, and then turn it so it's oriented properly."
"To the robot's dexterity and sensing ability, the system adds reasoning ability - so that the robot can acquire and orient parts by itself, instead of relying on a dedicated mechanical feeder," said Carlisle.
In a paper presented this month at the IEEE International Conference on Robotics and Automation in San Diego, Goldberg, Carlisle and their collaborators - Jeff Wiegley, an IRIS graduate student, and Anil Rao, of the University of Utrecht, the Netherlands - offered a blueprint for the system they're designing for the production floor.
The team will use a recently announced $250,000 grant from the California Trade and Commerce Agency's Office of Competitive Technology to realize this industrial system.
In addition to Carlisle, John Craig, of Cupertino-based Silma Inc., will participate in the research effort. Major manufacturers, including General Motors, Motorola, AT&T, Eastman Kodak, McDonnell Douglas, AVEX Electronics and Feeder Systems Inc., have already expressed interest in the new feeder technology.
Using a video image to guide a robotic device to pick up and properly reorient an object is not as simple as it might seem.
The first step, Goldberg explained, is to program the computer to recognize, from a video image, all the stable positions a part can assume on the conveyor belt.
Each possible view is then tied to a sequence of robotic motions for reorienting the part. Since the robot has only a simple two-fingered gripper, working out the motion sequence is no trivial task.
"Pick up a matchbook just with your forefinger and thumb," Goldberg suggested. "Now try to put it back on the desk with the reverse side facing up - without using another finger to help turn it over."
To solve this problem, Goldberg has patented a design for a programmable gripper using a translational bearing system; more recently, he and his co-workers have designed a rotational bearing system. "Our approach has been to address the problem using a rigorous analysis of geometry and mechanics."
Goldberg ticked off the system's advantages over existing technology:
The prototype, built in an IRIS lab, can feed a part every five seconds. In the larger-scale research just begun, Goldberg and his co-workers aim to achieve much higher speeds.
They also want to refine their computer algorithms so that they can take an image of a given part in a computer-aided design display and automatically create a set of gripper moves that will adjust the part to any desired orientation.
Goldberg noted that initial installation costs will be somewhat higher for the self-feeding robot and that conventional parts feeders will still be preferable for feeding high-volume, low-value parts, such as screws.
But with mechanical parts feeders currently accounting for 30 percent of the setup costs and causing more than 50 percent of the on-line failures in automated manufacturing, according to standard sources, there is room for improvement in existing methods.
"We don't want to make claims about our system's performance before we have thoroughly tested it in industrial applications." Goldberg said. "Parts feeding has long been a bottleneck in robot assembly, and manufacturers are rightly skeptical about claims of a magic bullet. Nevertheless, we believe our approach has great promise for improving California's competitiveness."