Just Roll With It

The design of every robot is very carefully tailored to its intended use cases. A robot meant for climbing its way through rough terrain is not going to win any races on level ground, and a speedy wheeled bot is not going to get very far through a field of boulders. So when versatility is needed, engineers have had to come up with some pretty wild designs, like this transforming car/drone or this flying wheel.

But researchers at the Hong Kong University of Science and Technology have shown that robot designs do not have to be quite that wild to be more adaptive. They have essentially taken a traditional quadrupedal robot dog and stuck some wheels at the ends of the legs. OK, so there is more to it than just slapping on wheels, but that is the main thrust of it. And this change has an outsized effect — the robot can zip around fast and efficiently on wheels, then when the going gets tough, it can climb its way through rough terrain or up stairs using its legs.

This is not the first time a quadrupedal robot has been fitted with a set of roller skates. ANYmal, for instance, has been skating along for years. But the team’s new robot, called FLORES, has some important tweaks that make its dual-mode operation more effective than designs of the past.

The team’s key insight was that wheels and legs should not just be bolted together, but integrated into a cohesive system with carefully rethought joints. FLORES’s most distinctive feature is its front-leg configuration. Instead of using a standard hip-roll joint that moves the legs outward, the designers swapped in a hip-yaw joint that rotates them horizontally. This gives the robot much greater steering ability on wheels, allowing it to change direction more fluidly without lifting its legs. The rear legs, meanwhile, retain a more conventional structure, letting the robot maintain stability and agility when it needs to walk.

This design means that FLORES does not have to stop and shift modes in the clumsy way earlier robots often did. Instead, it can glide along smooth surfaces like a wheeled vehicle, then adapt to rough terrain by walking or climbing. In the future, the robot may even be able to switch to a bipedal mode for tricky situations, such as crossing a narrow bridge.

The team ran a series of tests against other wheel-legged robots to find out how theirs would stack up. FLORES proved to be significantly more energy-efficient, using just 30 percent of the power of comparable robots when moving in a straight line and 35 percent when making turns. That translates to longer runtimes and greater practicality in real-world missions, whether in delivery, rescue, or collaborative work with humans.

Aside from the hardware upgrades, the improved performance was also made possible by a reinforcement learning controller customized for FLORES. Their algorithm adapts the so-called Hybrid Internal Model framework (a method for controlling legged locomotion) to the robot’s unusual configuration, rewarding behaviors that make the most of its steering and terrain-handling strengths. The result is a system that can generate smooth, adaptive strategies for switching between rolling and walking without human micromanagement.

Wheel-legged robots are still a relatively new category, but FLORES demonstrates that small design changes can unlock big performance gains. Rather than creating ever-stranger hybrid machines, sometimes the smartest step forward is a carefully considered reconfiguration of the basics.