When designing a mobility system for a robot, the goal is usually to come up with one single system that allows your robot to do everything that you might conceivably need it to do, whether that’s walking, running, rolling, swimming, or some combination of those things. This is not at all how humans do it, though: If humans followed the robot model, we’d be walking around wearing some sort of horrific combination of sneakers, hiking boots, roller skates, skis, and flippers on our feet. Instead, we do the sensible thing, and optimize our mobility system for different situations by putting on different pairs of shoes.
At ICRA, researchers from Georgia Tech demonstrated how this shoe swapping could be applied to robots. They haven’t just come up with a robot that can use “swappable propulsors”—as they call the robot’s shoes—but crucially, they’ve managed to get it to the swapping all by itself with a cute little robot arm.
Nifty, right? The robot’s shoes, er, propulsors, fit snugly into t-shaped slots on the wheels, and stay secure through a combination of geometric orientation and permanent magnets. This results in a fairly simple attachment system with high holding force but low detachment force as long as the manipulator jiggers the shoes in the right way. It’s all open loop for now, and it does take a while—in real time, swapping a single propulsor takes about 13 seconds.
Even though the propulsor swapping capability does require the robot to carry the propulsors themselves around, and it means that it has to carry a fairly high DoF manipulator around as well, the manipulator at least can be used for all kinds of other useful things. Many mobile robots have manipulators of one sort or another already, although they’re usually intended for world interaction rather than self-modification. With some adjustments to structure or degrees of freedom, mobile manipulators could potentially leverage swappable propulsors as well.
In case you’re wondering whether this additional complexity is all worthwhile, in the sense that a robot with permanent wheel-legs can do everything that this robot does without needing to worry about an arm or propulsor swapping, it turns out that it makes a substantial difference to efficiency. In its wheeled configuration on flat concrete, the robot had a cost of transport of 0.97, which the researchers say “represents a roughly three-fold decrease when compared to the legged results on concrete.” And of course the idea is that eventually, the robot will be able to handle a much wider variety of terrain, thanks to an on-board stockpile of different kinds of propulsors.
The robot uses a manipulator mounted on its back to retrieve the propulsors from a compartment and attach them to its wheels. Photos: Georgia Tech
For more details, we connected with first author Raymond Kim via email.
IEEE Spectrum: Humans change shoes to do different things all the time—why do you think this hasn’t been applied to robots before?
Raymond Kim: In our view, there are two reasons for this. First, to date, most vehicle-mounted manipulators have been primarily designed to sense and interact with the external world rather than the robot. Therefore, vehicle-mounted manipulators may not be able to access all parts of the robot or sense interactions between the arm and the vehicle body. Second, locomotion involves relatively high forces between the propulsion system and the ground. Vehicle-mounted manipulators have historically been lightweight in order to minimize size, mass, and power consumption. As a result, such manipulators cannot impose large forces. Therefore, any swappable propulsor must be both capable of bearing large locomotive loads and also easily adapted with low manipulation forces. These two requirements are often at odds with each other, which creates a challenging design problem. Our ICRA presentation had a failure video that illustrated what happens when the design is not sufficiently robust.
How much autonomy is there in the system right now?
Currently, autonomy is limited to the trajectory tracking of the manipulator during the process of changing shoes/propulsors. We initiate the change of shoe based on human command and the shoe changing operation is a scripted trajectory. For a fully autonomous version, we would need a path-planning algorithm that is able to identify terrain in order to determine when to adapt. This could be done with onboard sensing or a pre-loaded map.
Is this concept primarily useful for modifying rotary motors, or could it have benefits for other kinds of mobility systems as well?
We envision that this concept can be applied to a broad range of locomotion systems. While we have focused on rotary actuators because of their common use, we imagine changing the end-effector on a linear actuator in a similar manner. Also, these methods could be used to modify passive components such as adding a tail to the back of a robot, a plow to the front, or redistributing the mass of the system.
Currently the robot’s propulsors are designed for rough terrain, but the researchers are exploring different shapes that can help with mobility in snow, sand, and water. Photo: Georgia Tech
What other propulsors do you think your robot might benefit from?
We are very excited to explore a broad range of propulsors. For terrestrial locomotion, we think more tailored adaptations for snow or sand would be valuable. These may involve modifying the wheels by adding spikes or paddles. Additionally, we were originally motivated by naval operations. Navy personnel can swim to shore using flippers and then switch to boots to operate on land. This switch can dramatically improve locomotive efficiency. Imagine trying to swim in boots, or climbing stairs with flippers! We are looking forward to similar designs that switch between fins and wheels/legs for amphibious behaviors.
What are you working on next?
Our immediate focus is on improving the performance of our existing ground vehicle. We are adding sensing capability to the arm so that swapping propulsors can be performed faster and with greater robustness. In addition, we are looking to tailor motion planning algorithms with the unique features of our vehicle. Finally, we are interested in examining other types of adaptations. This can involve swappable propulsors or other changes to the vehicle properties. Manipulation creates a great deal of flexibility, and we are broadly interested in how new types of vehicles can be designed to take advantage of manipulation based adaptation.
“Using Manipulation to Enable Adaptive Ground Mobility,” by Raymond Kim, Alex Debate, Stephen Balakirsky, and Anirban Mazumdar from Georgia Tech, was presented at ICRA 2020.
[ Georgia Tech ]