Video 1 Video 2
Matt Bunting has been fascinated with robots since he was a young boy. He built his first one with Legos and a web cam. He used it to chase the cat. He sought out engineering and electronics projects in school and kept experimenting. As an UA junior, he took a cognitive robotics class and built a hexapod that could teach itself to walk and even reconfigure itself if damaged.
"Wheeled robots are simple and boring," Bunting said. "Hexapods are much more fun." Each leg of the hexapod has three motors for a total of 18 motors - just for the legs. He'd wanted to build a hexapod for years. "I grabbed all these spare parts I'd been collecting since I was very young. It was a hodgepodge of things. The legs were mismatched. Some were strong, some were weak."
But the concept was most impressive. His teacher - Anthony Lewis, Associate Professor of Electrical and Computer Engineering - gave him an A and offered him a job in the UA's Robotics and Neural Systems Laboratory (RNSL). "The robot was pretty amazing," Dr. Lewis said. "In addition to being more agile and elegant than other hexapods I have seen, what separates Matt's robot from all others is that it has a visual guidance mode. The robot can process visual information with insect-like neural processing and use that information to guide its legs over rough terrain. Further, the robot can relearn how to walk if a motor fails or a leg breaks."
"This kind of capability will definitely be incorporated in sophisticated space applications, where there is no technician around to make repairs, and where guidance of every foot placement is too tedious. Advanced versions of this robot could search in collapsed buildings for survivors, like after an earthquake," Dr. Lewis said.
Over the summer, Bunting decided to give his hexapod a makeover, using a 3-D printer to fabricate some of the plastic components, new motors, a web cam and an Intel Atom Processor as the robot's "brain". Bunting said, "I didn't need a display, a keyboard or a mouse. All I wanted to use it for was computation. It was really easy to integrate. I took a video of it and posted it on YouTube. Two days later this guy from Intel writes a comment. They were really interested in my robot because I didn't use the processor as a traditional computer. He asked me to contact him through Twitter - all the guys at Intel use Twitter. So I created a Twitter account. We talked and planned for him to come see the hexapod and meet with me and my professor," Bunting said.
"I let him control it. He really liked it." Intel asked him to build two hexapods. "I used all the best parts I could find. It was so much nicer than the one I made over the summer. Intel kept one to use for demonstrations and conferences and the other one was a gift to me so I could have it for my research."
Bunting doesn't program the hexapod to walk. "I wanted the hexapod to figure it out." A camera mounted on the front takes successive images of each stance, which are processed to help the robot determine if it is moving forward, sideways, backward or tilting. "This is reinforcement learning," he said.
Bunting continues to refine his hexapod. "It's a never-ending project. It's always changing." Next he wants to make a miniature - one that would fit in the palm of your hand. The current one stands eight inches tall and measures 14 inches by 17 inches.
The 2010 graduate plans to stay at the UA for an advanced degree, continuing to study electrical engineering and do research in Dr. Lewis' lab. "I think he's going to end up being a really outstanding robot designer," Dr. Lewis said.
A different YouTube video.