Do Robot Swarms Work Like Brains?
A new study by the Johns Hopkins University School of Medicine and Applied Physics Laboratory will combine research into navigational planning in brains with autonomous robotic swarms to drive advances in both fields. In this illustration, based on research into rat brain activity, measurements of brain activity can be translated into mapping instructions. Within the large circle on the right, the upper left circular image shows a neuron firing rate as the animal moved through space; the image at upper right shows theta rhythms from the hippocampus during the same movement. These are combined to create the bottom image, which shows phase attractors — a key element to creating a map based on those movements. The large circular image on the left shows, in red and orange, the predicted position of the rat in space from the analysis on the right; the blue line shows the rat’s actual path.
Credit: Joseph Monaco/Johns Hopkins School of Medicine
New Johns Hopkins Study Explores Navigation Similarities Between the Mind and Robot Swarms
Mon, 10/01/2018 - 13:19
A new study by the Johns Hopkins University School of Medicine (SOM) and Applied Physics Laboratory (APL) will combine research into navigational planning in brains with autonomous robotic swarms to drive advances in both fields.
“This is a very exciting new project from the perspective of a theorist and computational neuroscientist,” said Kechen Zhang, associate professor at SOM, and the study’s principal investigator. “We propose that individual robots in a group can be thought of as neurons in an animal’s brain. They interact with one another to form dynamic patterns that collectively signal locations in space and time, much in the same way brain rhythms do.”
The study (abstract) will use new information and discoveries about how the brain allows an animal to navigate and change its routes while moving — called dynamic replanning — to improve swarming algorithms to the point that groups of robots will automatically adapt to changes in the environment in the same way that a rat knows which detour to take around an unexpected obstacle. In turn, the neuroscientists will examine the replanning behaviors of drone swarms to evaluate their models of how the rodent brain dynamically replans paths, which will lead to new avenues for neuroscience research.
Zhang will work with a multidisciplinary team of co-principal investigators from APL, including applied mathematician Kevin Schultz, neuroscientist Grace Hwang, robotics researcher Robert Chalmers and STEM program manager Dwight Carr. Joseph Monaco, a postdoctoral biomedical engineering fellow at SOM, is a theoretical modeler on the study. The project was one of 18 funded in September by the National Science Foundation (NSF) to conduct innovative research on neural and cognitive systems.
In this movie, the rodent is stationary at a position on the left side of the lower video. A series of sharp waves (top video) are emitted by the rodent’s brain that generate sequences of spatially decoded locations. These decoded locations are represented by the traveling “hot spot” in the lower video, and show how the rat targets a goal or location to which it is considering moving. The goal location is shown as a square grid on the right side of the lower image.
Credit: Joseph Monaco/Johns Hopkins School of Medicine
The study’s genesis began in November 2017, when APL’s Hwang came upon a poster by Zhang and Monaco, based on data from a UCLA collaborator, at the Society for Neuroscience conference in Washington, D.C. The poster described their work with a type of neuron they had discovered and named a “phaser cell,” surprisingly found outside the hippocampus in a deep, centrally located region of the brain. Zhang and Monaco theorized that these neurons are critical to how the brain maintains its sense of location during spatial navigation; an animal’s brain uses spike timing patterns in the electrical activity of these neurons (called phase) to determine its specific location in space. Hwang, a neuroscientist with an interest in brain-inspired robotics, noted that unlike other neurons used in spatial perception, phaser cells create a phase map of the environment that is absolute (oriented to fixed directions such as north and south), not relative (i.e., “on my left”).
“That was the big ‘a-ha’ moment for me,” said Hwang, “because both the phaser cells and some robotic control algorithms use phase to determine location. For the first time, I saw a way to create a cognitive map of the entire environment using phase.”
Soon after that meeting, Hwang — who works in the Laboratory’s Intelligent Systems Center — was awarded APL internal research and development funding for an idea to develop brain-inspired small robots for GPS-denied areas. That proposal grew to include applied mathematicians Clare Lau and Schultz, and robotics researchers Chalmers and Bryanna Yeh. “We have neuroscientists here in the Intelligent Systems Center who collaborate with artificial intelligence researchers and roboticists,” said Chalmers. “And APL has an amazing hardware history in swarming vehicles. This is exactly the kind of cross-discipline collaboration the Center is designed to encourage.”
In early 2018, the project was awarded expanded R&D funding; encouraging simulations produced by Hwang and Schultz from that study led Zhang and the APL team to submit their proposal, “Spatial Intelligence for Swarms Based on Hippocampal Dynamics,” for an NSF grant. The key realization of the NSF proposal was the need to include an additional emergent phenomenon that occurs in the hippocampus, called sharp waves, that has been hypothesized to contribute to navigational planning in mammals. (Additionally, a new, related project led by Hwang has also received additional R&D funding from APL.)
The project also includes development of “Swarming Powered by Neuroscience,” a 16-hour micro-seminar for high school students that will be designed and operated by APL’s STEM Office. Robot swarms will be used to teach students about neuroscience, and the content will eventually be transformed into a course at APL’s STEM Academy.
Swarms of robotic agents such as these, with each agent representing a neuron, will be tested in realistic simulation environments. APL will use neuroscience research and modeling from the School of Medicine to mathematically analyze phase-organized swarming and dynamic replanning using matrix manifold methods.
Credit: Bryanna Yeh/Johns Hopkins Applied Physics Laboratory
The team envisions that this new approach to navigation will enable the kinds of tasks that society will be increasingly asking robots to perform — disaster relief and search and rescue, in addition to research and defense applications. These tasks require improved and more intelligent spatial coordination among many robots spread over large geographical areas. The team hopes that their research will create a revolutionary algorithmic framework for autonomous behaviors in swarming, as well as informing theoretical advances in understanding the brain. “What we hope to achieve is a more elegant way of expressing swarming behaviors,” Chalmers said.
“I have been at Hopkins for over 15 years, and this is my first collaboration with APL at the intersection of neuroscience and robotics,” said Zhang. “I look forward to a fruitful collaboration at the interplay of neuroscience, engineering and robotics that will find brain-inspired solutions to controlling distributed groups of robotic agents.”
Read the paper on phaser cells by Joseph Monaco , Rose De Guzman, Hugh Blair, and Kechen Zhang in PLOS Computational Biology: Spatial Synchronization Codes from Coupled Rate-Phase Neurons
The Applied Physics Laboratory, a not-for-profit division of The Johns Hopkins University, meets critical national challenges through the innovative application of science and technology. For more information, visit www.jhuapl.edu.