Most legged robots are easily identifiable as such, because we all know what legs look like: they look like legs. Maybe human legs, maybe mammal legs, maybe bird legs, but legs are legs. Where things start to get interesting is when legged robots manifest designs that aren’t (usually) found in nature, like with RHex, which has six springy wheely leggy things that allow it to do some incredible acrobatics.
At Carnegie Mellon University, Simon Kalouche just wrapped up his masters thesis in which he describes the development of a brand new design for “legs capable of dexterous walking, running, and most significantly, explosive omni-directional jumping and actively compliant landing.” That’s the kind of thing we like to hear.
Kalouche has developed the GOAT leg, which is obviously an abbreviation for “Gearless Omni-directional Acceleration-vectoring Topology.” The design of the leg has very little to do with the physical structure of a goat’s leg, but Kalouche was directly inspired by mountain goats, which display nearly supernatural locomotion on sheer cliffs. Rather than try to mimic the biology, Kalouche focused on ways of giving a robotic leg goat-like abilities, which resulted in a novel and highly dynamic parallel system driven by a trio of electric motors. The GOAT leg is highly dynamic, and did we mention that it’s highly dynamic? Because it’s highly dynamic:
Extra-terrestrial landscapes or a collapsed rubble environment, ubiquitous to war and disaster zones, will contain regions of highly rugged yet relatively level ground. In these environments using high bandwidth virtual compliance, made possible by low impedance actuators, will allow the robot’s legs to actively conform to the terrain producing a more efficient and swift mode of locomotion as compared to a statically stable crawling gait which requires accurate terrain mapping and explicit foot step planning.
Alternatively if the terrain is both sloped and rugged it may be ideal to crawl or climb slowly using precise footholds made possible by dexterous limbs with a large workspace. Likewise, unstructured or collapsed disaster environments often contain local discontinuities (e.g. cavities, pits, ditches, curbs, obstructions, large local elevation changes relative to the robots leg length, etc.) in the robot’s path.
To achieve mobility over such a broad set of terrain topographies - spanning structured and unstructured environments - an ideal robot will employ both static, highly stable motions (e.g. dexterous crawling, climbing, walking), as well as highly dynamic agility maneuvers (e.g. leaping, inertial reorientation, controlled landing; running; etc.) to optimally traverse the terrain at hand. Therefore, a capable legged robot must be both dexterous, for precise footstep placement, and dynamic, for running and jumping when obstacles are insurmountable by static gaits alone.
The GOAT leg uses a quasi-direct-drive actuator driven by a T-Motor U10 with a single-stage Matex 1:7 planetary gear.Image: Simon Kalouche/CMU
With a very few very specific exceptions, legged robots are not capable of dynamic motions. Humanoids are particularly bad at things like running and jumping, and it’s only very recently that quadruped robots have begun to display basic competence at the kinds of locomotion skills that humans (and especially animals) take for granted. Even so, these robots aren’t really cut out to do what legs are great at doing, which is traversing the estimated 50 percent of the planet that you can’t access with tracks or wheels. It’s these specific situations that the GOAT leg is intended to tackle:
- Highly unstructured terrains, that are often impassible to wheeled vehicles and currently existing legged robots.
- Obstacles of large and steep variations in ground elevation (relative to the robots height) creating discontinuous paths for walking (i.e. theres a limit to the slope of a terrain that can be walked over which is around 45 degrees for humans).
- Pits, holes, ditches and local cavities which also create discontinuities in a walking path.
- Tight and compact spaces where turning or reorienting to then walk or jump is not possible or ideal.
- Long distance missions over diverse terrains where different gaits or modes of locomotion can improve both mobility, efficiency, and longevity.
One of the things that sets the GOAT leg apart from other robotic legs is that it is not constrained to planar motions, making it much easier to move in different directions without having to reorient. The GOAT leg is also mechanically robust, exhibits virtual compliance (thanks to a modified impedance controller), and with its three hip-mounted electric motors, provides a high amount of distributed torque while keeping the inertia of the leg itself low.
Some state-of-the-art dynamic legged robots: 1) Boston Dynamics’ SpotMini; 2) UPenn/Ghost Robotics’ Minitaur; 3) ATRIAS; 4) RHex and Canid; 5) StarlETH; 6) Boston Dynamics’ Spot; 7) MIT’s Cheetah; 8) Boston Dynamics’ Atlas.Image: Simon Kalouche/CMU
It’s tempting to compare the GOAT leg to other designs, such as those of MIT’s Cheetah, a quadruped also powered by electric motors (other state-of-the-art legged robots mentioned by Kalouche in his thesis are shown above). Doing so is like comparing an actual goat to an actual cheetah: the goat’s not going to win any races, but it’ll at least be competitive, while the cheetah would likely be utterly hopeless at negotiating the kind of terrain that a goat calls home. Kalouche summarizes:
While the GOAT leg will most likely never be faster than the very impressive speeds achieved by the MIT Cheetah in running along a straight line, the design of GOAT sacrifices 1D top speed for the potential to run and jump in all 3 dimensions with more agility than the MIT Cheetah and all other dynamic legged robots to this date.
A CAD rendering of a robot design with four GOAT legs.Image: Simon Kalouche/CMU
Despite all the comparisons that we’re making to existing legged robots, you’ve certainly noticed by now that the GOAT leg is just one single leg that can’t actually stand up by itself. Kalouche is still experimenting with the leg hardware, and one of the next steps is mounting the custom electronics, motor controllers, and power system on-board the robot. Right after that he plans to remove the leg from the test rig and do experiments in fully unconstrained 3D space.
Beyond that, Kalouche has plans to integrate the GOAT leg into a variety of different robot morphologies, including monopods, bipeds, tripods, and that quadruped in the rendering above.
Because the leg is capable of delivering omni-directional force, a biped or quadruped using the leg topology will most likely behave unlike any other legged robots or legged animals. Instantaneous direction changes while running have the potential to create very unique mobility for future robots that has been unseen to this point in time.
You can read Simon Kalouche’s entire masters thesis here.