Of Rats and Robots
Researchers hope to replicate the rodents’ location-finding sense, then install it in robots| From Explorations | By Art Jahnke
Michael Hasselmo and his team hope to enable robots with both visual information and self-motion information from sensors that track acceleration and velocity. Photo by Kalman Zabarsky
The problem with robots, says Michael Hasselmo, is that they don’t always know when they’re someplace they’ve been before.
“They have been working to solve this problem over the past 20 years,” says Hasselmo, a College of Arts & Sciences professor of psychology and director of the Computational Neurophysiology Laboratory. “But often when you put a robot in a big square hallway, it will go down the hall, mapping the walls, building a representation of distance, then come all the way around and in many cases won’t know when it’s at the starting point.”
Rats, on the other hand, have the starting point problem licked. Or perhaps sniffed. Hasselmo isn’t sure.
“A rat can run around an alley just like the robot did,” he says. “He can find some scraps in an alley, then go away and come back and know that he’s looking in the same spot for more scraps. Rats are way better at this than robots.”
There are several possible reasons why rats are so much better than robots, says Hasselmo, and those reasons include noses, eyes, ears, and whiskers. The brain expert is uncertain of the hierarchy of senses used by rats to determine location, but he knows that they also use their vestibular sense of movement for something called path integration, an ability to understand how far they’ve moved in which direction.
Hasselmo explains path integration using the example of someone waking up in dark hotel room and knowing, as they walk around, if they are close to the TV or the bathroom.
“You know your location just by how far you step,” he says. “And if you turn your head and it’s complete darkness, you still know the direction you face. This involves integrating a velocity signal and your sense of direction. If you’re moving through an environment and you have sensors telling you what your direction and speed are at every point in time, you can integrate that information and get position. You can keep track of location by summing up your movements—that is, by integrating your path through the environment.”
Many people would like to have a better understanding of how this mechanism for navigation works. The U.S. Office of Naval Research, for example, which would like to make robots that could figure out and communicate where they are within a building, is sufficiently interested in getting a better understanding that it has funded a $7.5 million Multidisciplinary University Research Initiative (MURI), with Hasselmo, the principal investigator. Hasselmo, who is the associate director of the Center for Memory and Brain, is working with center director Howard Eichenbaum, a CAS professor of psychology, and Chantal Stern, a CAS professor of psychology and director of the Cognitive Neuroimaging Laboratory. The team, pulled together by Hasselmo, also includes three researchers at MIT, one at the University of Texas at Austin, and one at University College in London.
Biologically, says Hasselmo, it’s easy to see the benefit of accurate tracking of location. “If a rat is running down an alley and it finds a nice big garbage can filled with leftover Chinese food, it’s very useful to be able to find that location again,” he says. “And conversely, if he is walking down an alley and a cat comes out and chases him, he might want to remember where exactly he got chased.”
Neurons in rat brains fire when a rat is where it was when the very same neuron last fired. And they do it consistently.
Robots, on the other hand, which have little interest in Chinese food, rely on researchers to endow them with navigational mechanisms. They are struggling with a problem referred to as SLAM, or simultaneous localization and mapping. “A robot can come to a spot where he thinks he started,” says Hasselmo, “and it’s many yards away.”
Across the Charles River at MIT, team member Nicholas Roy, an associate professor of aeronautics and astronautics, is focusing on the robotics side of the puzzle. Roy is working with flying machines that look to most people like little helicopters with four propellers. He has outfitted the robots with laser sensors that pick up the distances to walls and with pairs of video cameras that can pick up visual signals and determine depth of field. “They’re very cool,” says Hasselmo, “but they can’t keep track of their location as well as rats do.”
In order for the team to build robots that do what rats do, the members have to first figure out how rats do it. That’s where Hasselmo, whose expertise is memory-guided behavior, comes in. For years, he has been studying a subset of brain cells, called grid cells, in rats’ brains as his researchers track the location of the rats using tiny diodes on the rats’ head. When a grid cell fires, as they do frequently, the location of the rat is recorded. Previous researchers found that the same neurons fire when the rat is in the same place as it was when they last fired, and they do it consistently.
“We toss tiny bits of crushed cereal on the surface, and the rat does what rats love to do—forage for food,” he says. “And we spread the food out so the rat keeps wandering around. Every time a cell would spike, it would mark the location of the rat, so it looks as if the rat is coding his location in the environment.”
Hasselmo says you can run a rat for 20 minutes in a particular field—typically, in his experiments, a field no larger than one square meter—remove it from the field for an hour, put it back, and the neurons still fire in the same locations. What tells the cells to fire remains a mystery he would like to solve.
Other researchers have eliminated other sensory input as the key to solving the mystery. They’ve deprived rats of auditory input and input from whiskers, and the rats can still find their location. “The standard view is that the rat is doing path integration,” says Hasselmo. “It’s keeping track of its movement and its head direction with its vestibular system.”
His effort to create models of grid cell firings has revealed that rats have grid cells that respond on different scales: some fire when the rat moves 40 centimeters, whereas others may fire when it moves about 80 centimeters. Others seem to fire at up to 10-meter intervals. Hasselmo says the different spatial frequencies arise from oscillations of brain activity over time. He believes that the rat brain relies on oscillatory interference to integrate speed and direction, thereby yielding a sense of position. Working with a basic oscillatory interference model, the team has been able to guide a wheeled robot. “It works on a pretty simple level,” he says. “It can find its way from a preassigned starting location to a preassigned goal location.”
When it comes to recognizing when they are passing through a previously visited location, the robots still have a long way to go, so to speak. Hasselmo hopes to use visual input to inform the robot of its starting point, then enable the robot to track its movements with a combination of visual information and self-motion information from sensors that detect acceleration and velocity.
Ultimately, he says, the goals of the MURI grant include understanding how rats keep track of their location and applying that mechanism to robots, so robots can work as a sidekick to soldiers searching dangerous buildings and alleyways. Ideally, the robots will navigate strange spaces and report back to their human colleagues on the exact locations of bombs and other dangers. But not Chinese food.