Building more stable four-legged robots

A biologist turned roboticist takes a closer look at dog gaits to help design better movements for four-legged robots.

Dog climbs over fallen tree Enlarge
Dogs can handle uneven ground very gracefully. Roboticists want to know their secrets. Credit: Kelly O’Sullivan, Michigan Engineering Communications & Marketing

Research into the gait of dogs may lead to improved design of quadruped robots and how we control their movement. Shai Revzen, a biologist turned roboticist, brings a unique perspective to the study of animals, one that’s beginning to be heard by the biological community as well.

“The research is shaking up both disciplines,” said Revzen. “giving biology a new tool to look at things, and giving robotics a sense of benefit from what we know in nature in a direct, simple way that’s not just naively copying what you see.” Revzen is an assistant professor of Electrical Engineering and Computer Science as well as Ecology and Evolutionary Biology.

In a collaboration with members of the Royal Veterinary College and other roboticists, Revzen sought to discover which of a dog’s many gaits it would favor when encountering disturbances in its environment, such as bumpy turf.

A traditional biologist would attempt to express the dogs’ movements with gait diagrams, which show when the legs touch the ground. Much richer information can be found by looking at where all the legs are in respect to their cyclical motions and each other.

When a dog trots, for example, their diagonal front and back legs rise and fall in a rhythmic pattern. In most quadrupeds the motion of an individual leg is not that different in gaits such as a walk, an amble, a pace, a trot, a canter, or a gallop. It’s the relative position of all the legs to each other that determines the gait.

It turns out that the dogs varied their gait between a walk and a trot depending on the bumpiness of the ground.

A trot is considered a dog’s most efficient gait, and the research confirmed that a dog is most stable trotting with respect to a longitudinal axis of movement. Perhaps counterintuitively, the dogs would drift to that gait when encountering perturbations in the environment, rather than slowing down immediately to a walk, for example.

What Revzen discovered in this research suggests that quadruped animals, like dogs, will attempt to move in a way that is longitudinally stable. This movement can be expressed mathematically. By applying the math to a quadruped robot, you might get a robot that can adapt more naturally to its surroundings. When speed is of the essence, such as in a search and rescue scenario, that extra time may become critical to a successful outcome.

Also – what if something happens to one of the four legs? Revzen believes by looking closely at what animals do and applying a mathematical model recovering the cyclic relationships between the legs, he’ll be able to program a robot to recover much more quickly than what is currently possible with other methods.

His ultimate goal is to formulate a theory of robust locomotion that can capture what’s happening in nature, in order to apply it to a field like robotics.

The research is described in the paper Longitudinal quasi-static stability predicts changes in dog gait on rough terrain, which has appeared online and will appear in a future publication of the Journal of Experimental Biology. The paper is co-authored by Simon Wilshin, Michelle A. Reeve, G. Clark Haynes, Shai Revzen, Daniel E. Koditschek, and Andrew J. Spence. Wilshin and Reeve are biologists at the Royal Veterinary College in London, England. Haynes is at Carnegie Mellon University, Koditschek at the University of Pennsylvania, and Spence is at Temple University.