By Wayne Gillam / UW ECE News
UW ECE Associate Professor Sam Burden is part of a multi-institutional research team that examined why walking, running, and jumping are challenging tasks for robots while the same activities appear to be relatively easy for humans and other animals. The team published their findings in a recent issue of the journal Science Robotics. Photo by Nicolas Thomas / Unsplash
Over the last few years, millions of people have watched videos of robots walking, running, and jumping with breathtaking power, agility, and speed. However, what many people don’t realize is that these videos are carefully choreographed and take place in tightly controlled environments. In the real world, outside of those controls, legged robots still have a long way to go to match what humans and other animals can do.
It turns out that walking, running, and jumping, or “legged locomotion,” as it’s known in engineering circles, is surprisingly difficult for robots, especially when it comes to achieving dynamic mobility in an uncontrolled environment. Digital, programmable robots have been around for decades now, but compared to animals, their skill at legged locomotion in the real world is barely out of its infancy. That’s not so bad when one considers that animals have had millions of years to evolve and perfect their moves. It even takes a human toddler several years to learn how to walk, run, and jump. So, with those points in mind, perhaps it’s not quite as surprising that it’s taking scientists and engineers a long time to master this difficult skill set on behalf of robots. But why are walking, running, and jumping such challenging tasks for robots, when the same activities seem to be relatively easy for animals?
In a new paper titled, “Why animals can outrun robots,” which was recently published in the journal Science Robotics, a multidisciplinary, multi-institutional research team that included UW ECE Associate Professor Sam Burden examines in depth why this might be.
“If you look at a squirrel, for example, it’s amazing what they can do. And there’s just no comparison at any scale or any kind of modality for legged robots,” Burden said. “The point of this paper is to synthesize across biology and engineering what we know about the components and the whole systems involved and try to answer the question of why animals are so much better at legged locomotion than robots.”
UW ECE Associate Professor Sam Burden. Photo by Ryan Hoover / UW ECE
Burden’s collaborators included Max Donelan, a professor at Simon Fraser University in biomedical physiology and kinesiology; Kaushik Jayaram, an assistant professor in the Paul M. Rady Department of Mechanical Engineering at the University of Colorado Boulder; Simon Sponberg, the Dunn Family Associate Professor of Physics and Biological Sciences at the Georgia Institute of Technology; and Tom Libby, who was a Washington Research Foundation Fellow in Neuroengineering at UW ECE from 2017 to 2019 and is now a senior research engineer at SRI International.
Each researcher in the group explored one of the five engineering subsystems that make up robotic legged locomotion. Together, they dug deep into the scientific literature, investigating and analyzing why animals outperform robots at walking, running, and jumping, and they quantified the differences they found. Before this research, many scientists and engineers believed the main reason animals had a significant advantage over robots was that biological components were superior to engineered parts. But what the team discovered because of their extensive review, was that the opposite was true, and that the whole was far greater than the sum of its parts.
“The way things turned out is that, with only minor exceptions, the engineering subsystems outperform the biological equivalents — and sometimes radically outperform them,” Libby said in a recent press release from Simon Fraser University. “But also, what’s very, very clear is that, if you compare animals to robots at the whole system level, in terms of movement, animals are amazing. And robots have yet to catch up.”
Based on these findings and their intensive examination of engineering subsystems, the team identified in their paper fundamental obstacles that roboticists must overcome to bring robot legged locomotion up to par with humans and other animals. The team also highlighted promising research directions that hold transformative potential to help legged robots achieve animal-level performance.
Engineering subsystems, overcoming obstacles, and promising research directions
The team’s paper was comprehensive in its review of the scientific literature available on this topic. Their research began in 2013 and lasted over a decade, as group members worked on investigation and analysis of legged locomotion in between their other responsibilities.
“In the paper, we divide legged locomotion into five engineering subsystems and cover them all in depth,” Burden said. “Normally, analyzing any single one of these subsystems for either an animal or a robot could be an entire review paper by itself. It’s an ambitious and broad project.”
The five engineering subsystems the team explored were the power system used to store and deliver energy, the frame that provides support and leverage, actuators to modulate mechanical energy, sensors to perceive self and environment, and the control system, which transmits and transforms sensor and actuator signals. For each subsystem, the team compared, contrasted, and quantified differences between legged robots and animals. Burden said that the group wrote this paper primarily for roboticists but that they also wanted their findings to be accessible to biologists to encourage collaboration when tackling the tough problem of improving robotic legged locomotion.
“This paper is rigorously researched, and basically, we’re saying that if you want high performance and want to approach the capabilities of animals, what we need in robotics is an integrative approach. It is not the quality of robotic components that explains this wide performance gap, but rather, how they are put together into a unified whole.” — UW ECE Associate Professor Sam Burden
To that end, the team identified four fundamental obstacles they believe must be overcome to successfully integrate engineered components into more effective robotic systems. Those obstacles are a lack of quantitative metrics for evaluating the many dimensions of legged locomotion; the tradeoffs that arise when subsystems combine and the performance of one component potentially constrains the performance of another; the phenomenon of emergence, where the behavior of the whole system is different from, and irreducible to, the behavior of its component parts; and the very Harry Potter-sounding curse of dimensionality, which means that there is a mind-boggling array of possible component configurations roboticists can choose from when designing legged robots and very little guidance as to which will be the most effective.
To not leave scientists and engineers without paths to solutions, the researchers also identified several promising research directions. Those include systematic comparative studies of multiple animal species, which could reveal generalizable principles that could be applied to robotics; distributing energy, sensing, actuation, and control throughout robot frames, as animals do, which may enhance robustness and advance autonomy; bridging the “sim-to-real” gap with better computational models of robot interactions with the environment; continuing advances in materials used to build robotics, and systematically exploring tradeoffs with respect to multiple performance metrics at both component and system levels.
Overall, the research team emphasized that although further improvements to robotic components are beneficial, the greatest opportunity to improve the performance of legged robots is to make better use of existing parts, much like biological systems do. They advocated in the paper for a more integrated approach to engineering legged robots, taking cues and guidance from biology along the way.
Downstream impacts, ethical considerations, and looking ahead
Burden and his colleagues are optimistic that over the long term, when it comes to developing robots that can walk, run, and jump as well as or better than humans and other animals, the benefits will far outweigh the risks. Photo courtesy of the Animal Inspired Movement and Robotics Lab / University of Colorado Boulder
By developing a better understanding of the principles involved in legged locomotion for both animals and robots, Burden and his colleagues have moved the field of robotics closer toward a longstanding goal for engineers — creating robots that can walk, run, and jump as well as (and perhaps even better than) humans and other animals. There are many reasons why this is an important, worthy goal. Legged robots with robust agility could perform many useful, and even life-saving, tasks in environments that are hazardous for humans, such as cleaning up after natural and nuclear disasters, disarming bombs, or helping astronauts explore outer space. Principles learned from this robotic development could also be applied to advanced, bio-inspired devices, such as smart prosthetic limbs and exoskeletons. And the potential everyday applications are endless, including developing legged robots to clean the house, do yard work, and even care for the elderly. The automation of various tasks by legged robots across a vast range of industries also promises to substantially enrich the world economy.
But, of course, every powerful technology can be a double-edged sword, and there are some downsides to consider. Robotic automation could enrich the economy, but that will be at the cost of job loss for at least some humans. This could happen in large numbers and at such a rapid pace, it would be hard for society to adjust. The possible weaponization of legged robots also is a serious concern, and some manufacturers are calling on the robotics community and government leaders to take steps to ensure this doesn’t happen. Recently, some thought leaders have suggested that the fear of job loss from robotics is overblown; however, whether or not they are right still remains to be seen. In the meantime, roboticists, industry leaders, and government representatives are exploring different avenues for addressing these sorts of concerns, and that work is ongoing.
Burden and his colleagues are optimistic that over the long term, when it comes to developing robots that can walk, run, and jump as well as or better than humans and other animals, the benefits will far outweigh the risks.
“These are machines that could have a really big, positive impact on people’s lives, but they’re just not capable yet,” Burden said. “This paper is rigorously researched, and basically, we’re saying that if you want high performance and want to approach the capabilities of animals, what we need in robotics is an integrative approach. It is not the quality of robotic components that explains this wide performance gap, but rather, how they are put together into a unified whole.”
Learn more about this research by reading “Why animals can outrun robots” in Science Robotics. More information about UW ECE Associate Professor Sam Burden is available on his bio page.