A New Theory of Robotics?
Robots have come a long way in the past few decades, but we’re still a long way off from one that can move like animals and humans. To date, programming movement for robots uses instructions to individual mechanical parts to mimic muscle activity. The main challenge is that the number of small, coordinated muscle movements in walking requires an enormous number of instructions to program. In addition, these instructions are often not very good at accommodating for different surfaces or changing landscapes.
One way around this issue might be to focus less on “muscles” and more on neurons for creating the instructions of walking. This is the approach being taken in the lab of Francisco Valero-Cuevas, PhD, Professor of Biomedical Engineering at the University of Southern California. A recent
feature at Wired magazine details their construction of a robotic cat based on a network of artificial neurons.
The USC model uses reinforcement learning, which is a system whereby neurons of the spinal cord form networks on the basis of trial and error, using random firing of neurons until motion is produced. In this way, the need for an algorithm or complicated programming is eliminated. The cat, called Kleo, is a long way off from being able to land on its feet or use a litter box, but it might give us insight into new technologies that will help people with disabilities from spinal cord injury or motor neuron disease.
Less Neuronal Flexibility With Learning
One of the primary tasks of the brain is learning, but there’s still a lot we don’t know about what happens in the brain as learning occurs. Much of the past research examined changes at the level of individual neurons to explain learning. Newer research, however, has indicated that it is more insightful to examine larger populations of neurons during tasks to get a deeper insight into how the brain learns.
Using this principle, a
team of engineers and scientists collaborating between Carnegie Mellon and the University of Pittsburgh submitted rhesus monkeys to a learning task and obtained neural recordings to determine how the task affected neuron populations. Their study, led by Steven Chase, PhD, and Byron Yu, PhD, both associate professors of biomedical engineering at CMU,
was published in
Nature Neuroscience. Drs. Matthew Golub and Penn alumnus Aaron Batista were also coauthors.
Contrary to previous thinking, the authors found the brain is less flexible during learning tasks. In part, this lack of flexibility explains why certain tasks take a long time to learn. The authors state that it remains unclear whether the brain changes detected occur at the level of the cortex or subcortex, so additional research will be necessary.
Preventing Bad Science
Academic science remains largely an environment of publish or perish, and this pressure on scientists has unfortunately resulted in an increased incidence of academic fraud. One form of fraud is recycling old images from past publications of successful experiments while presenting the results of newer research.
Recognizing that data science could be used to detect such episodes of fraud, Konrad Kording, PhD, a Penn Integrates Knowledge (PIK) Professor with appointments in the Departments of Bioengineering and Neuroscience, and his collaborators developed an algorithm that can compare images across journal articles and detect whether images have been repeated across two articles, even if they have been resized, rotated, or cropped. They describe their technique in a paper recently published on the BioRxiv preprint server. Among the next moves the authors are considering is licensing the algorithm to academic publishers, with the caveat that the possibility of false positive accusations has not been eliminated.
People and Places
Congratulations go to Judy Cezeaux, PhD, who has been named Dean of the Arkansas Tech University College of Engineering and Applied Sciences. A biomedical engineer with degrees from Carnegie-Mellon and Rensselaer Polytechnic Institute, Dr. Cezeaux was most recently chair of the Department of Biomedical Engineering at Western New England University.
