Science of Learning Innovation and Control, Embodied





















Army Research Office Grant W911NF-18-1-0327, Multidisciplinary University Research Initiative on Science of Learning Innovation and Control, Embodied (SLICE).

About

Written by Sonia Roberts and Evan Lerner on behalf of the SLICE team

A squirrel sits on a platform, tail twitching, seeming to carefully consider its next move. Suddenly, it leaps! The mill wheel it lands on spins under it, threatening to throw it off, but quickly the squirrel climbs towards the center of the wheel and jumps to another, more solid platform. A piece of string connects this platform to the next -- no problem at all.

There are dozens of videos like this on YouTube. Squirrels are willing to work for their nuts and seeds, and in response, people try desperately to create obstacles that will prevent their backyard bird feeders from becoming squirrel smorgasbords. But with ingenuity and acrobatic prowess, the squirrels always seem to overcome.

In contrast, even the most advanced legged robots can be stymied by very minor obstacles like piles of rocks or sand. They have no capacity for inventing novel behavior at all: each new gait or maneuver must be programmed from scratch. And any robot is put to shame by the average cat, squirrel or gecko, all of which can quickly adapt complex bodily features with manipulations to traverse almost any obstacle whether familiar or completely novel.

We are a team of engineers and scientists aiming to imbue robots with this kind of “embodied intelligence,” developing bio-inspired designs that use limbs as sensors as well as actuators and learn new forms of locomotion based on interactions with their environment. Our five-year goal includes a physical manifestation of the new insights in the form of a parkouring mechanical “squirrel” that will serve as a new paradigm for robot design and behavior.

Animals’ bodies are exquisitely adapted to their habitats. Evolutionary pressure has resulted in bones, muscles and skin that automatically “solve” physical problems of force, sensation and energetics, allowing their brains to handle more abstract problems of placement, motivation, strategy and novelty.

“Animals are constantly performing both mechanical and mental processing of the information and energy flows at work in their environment. ‘Embodied intelligence' refers to the integrated physical problem solving and innovative capabilities that emerge from this interplay.”

For example, squirrels are at home whether on the ground, traversing tree trunks or leaping branch to branch. They use their bodies expertly to solve different physical computations in each new environment. On level ground, their bounding gaits can be stabilized in part by changing their body’s mass distribution. Running down a tree engages mechanisms in squirrels’ ankles that bias the claws to “find” appropriately located grips. A leap is executed in part by tuning leg muscles to the springiness of the branch in the selection of a trajectory. Moreover, squirrels are constantly inventive, deploying suitably modified versions of previously discovered maneuvers as the need arises.

Engineers barely know how to even formulate, much less achieve such synergy in robot designs and programs. Koditschek’s own Kod*Lab has previously developed a series of agile, bioinspired robots that use their limbs as sensors. The lab has produced a robotic kangaroo-rat, Jerboa, and a spin-off company, Ghost Robotics, which makes Minitaur, a quadrupedal robot that can open doors with a back handspring among other tricks. These robots’ ability to “feel” the world with their limbs gives them uncommonly effective movement over a variety of terrains and the ability to manipulate objects encountered within them.

To help coordinate efforts across the team and the different disciplines represented, our effort is organized into the following three Research Concentration Areas (RCA)

the math that describes the information and energy exchange between the robot and its environment; how spatial information is encoded acted upon in biological brains and bodies working togetherRCA 2.i: Animal Learning and InnovationRCA 2.ii: Physiological Studies in Manipulated Environment new forms of robot body design, material construction and behavioral control that can take better advantage of what the animals reveal and the mathematics explains about fine-grained sensing and control. RCA 3.i: Morphological Design, Computation RCA3.ii: Morphological Innovation, Control and Behavior

Sponsor and Team

The project is supported by the Army Research Office through the Multidisciplinary University Research Initiative, or MURI, Program funded by the Office of the Under Secretary of Defense for Research and Engineering. It is being led by Daniel E. Koditschek, Alfred Fitler Moore Professor of Electrical and Systems Engineering.

Koditschek will collaborate with Shu Yang, professor in the Department of Materials Science and Engineering, as well as Yuliy Baryshnikov of the University of Illinois Urbana-Champaign, Noah J. Cowan, expert in neuromotor control, and James J. Knierim , neurophysiology of Johns Hopkins University, and Robert J. Full , expert in biomechanics, and Lucia F. Jacobs, expert in animal cognition, of the University of California at Berkeley. The team has recently been joined by Cynthia R. Sung, Professor of Mechanical Engineering and Applied Mechanics, who brings her expertise in folded active structures to the application of materials science to robotics. The SLICE MURI promotes integrated study of the way the brain and nervous system interact with the muscles and skeleton in accommodating and influencing the environment.