News — When walking on the sidewalk, a person is able to avoid puddles, other walkers, and cracks in the pavement. It may seem intuitive – and that's because it is.
There’s actually a biological component that allows humans and other mammals to navigate our complex environments. Central Pattern Generators (CPG) are neural networks that produce rhythmic patterns of control signals for limbs using simple environmental cues. When we quickly step away to avoid something blocking our path, that’s our CPGs doing their job.
Rajkumar Kubendran, principal investigator and assistant professor of electrical and computer engineering of the University of Pittsburgh, received a $1,606,454 award from the to lead a two-year project to engineer these neural networks in robots. Feng Xiong, associate professor of electrical and computer engineering at Pitt, and M.P. Anantram, professor of photonics and nano devices at the University of Washington, will serve as co-principal investigators.
“While these networks are natural for us, there is currently no efficient way to replicate them using electronic devices and computers,” Kubendran explained. “Agile robots that can explore unknown and treacherous terrains have the potential to enable autonomous navigation for commercial transport, enhance disaster response during floods and earthquakes or to remote and unsafe areas like malfunctioning nuclear plants or space exploration.”
A Robot on the Move
Building bio-inspired robots isn’t something new, but modeling, designing and implementing neuromorphic networks with synapses and neurons inside miniature robots is a novel step forward – or, in the robot’s case, also backwards, left and right.
Neuromorphic engineering – computing inspired by the human brain – will be key to achieving efficient, adaptive sensorimotor control in these robots.
“We aim to demonstrate a fully functional quadropod or hexapod robot that can learn to move, using principles informed by neuroscience, leading to biomimetic sensorimotor control for energy-efficient locomotion, using learning algorithms running on bio-realistic neural networks, built by using semiconductor technology that expands beyond its normal limits,” Kubendran said.
These robots will be able to avoid a puddle just as intuitively as humans by constantly learning about movement. The group will take inspiration from neural circuitry found in biology that supports agile movement control. The team plans to incorporate non-linear temporal dynamics in mixed-feedback systems to build bio-inspired neural networks and implement them on scalable energy-efficient hardware.
To meet the challenging demands of neuromorphic engineering, the team is also developing the NeuRoBots educational consortium. The consortium will train a new generation of engineers and researchers through evidence-based best practices to prepare them for the rapidly evolving needs of the industry.
“The breadth of skill sets that are required to effectively train a new cadre of workforce in neuromorphic computing for robotics makes curriculum design and integration with existing frameworks challenging,” Kudendran said. “We need help to prepare our engineers for this changing technical environment.”
The project, “Bio-inspired sensorimotor control for robotic locomotion with neuromorphic architectures using beyond-CMOS materials and devices,” is set to begin in 2024 and is part of a larger, $45 million initiative by the NSF to invest in the future of semiconductors.
RSS