
Researchers Developed a BMX Robot That Learned to Bunny Hop on Its Own.
I don't know about you, but I'm pretty tired of the whole robot trend. It's cool to see them climb stairs, play soccer, or shoot hoops, but haven’t these scientists and engineers seen Terminator? Don’t they realize they’re contributing to our eventual downfall? I wish they would stop, but they never will. So if society insists on creating robots, they might as well be like this awesome bunny-hopping BMX stunt machine.
Researchers at the Robotics and AI Institute (RAI) have developed a robot capable of bunny hopping higher than I ever could on a bike as a teenager. Setting aside my bruised ego, it’s an impressive feat and, more importantly, quite entertaining. This creation is called an Ultra Mobility Vehicle (UMV), and it's not meant to take jobs or be used in military functions (though its technology will likely end up in those areas); it’s simply a dynamic BMX bike that performs tricks autonomously.
To enable the UMV to hop, land, and maintain balance on its own, RAI needed to create a new control unit that utilizes reinforcement learning—a type of machine learning where the software guiding the robot determines the most effective and efficient ways to interact with its surroundings. This process involves considerable trial and error, but RAI claims that reinforcement learning enables both the team and the robot to transcend the limits of normal direct control and unlock the full potential of the hardware. For instance, RAI applied reinforcement learning to a Boston Dynamics robot dog, which enhanced its speed from 1.6 meters per second upon initial testing to 5.2 meters per second.
The video below showcases the UMV in action. On its “seat” is the hefty control unit, connected to the bike via a flexible mechanical arm. Remarkably, this computer does not use a gyroscope or stabilization sensors—it’s just an ordinary bike with a large, heavy backpack, and it has learned to accelerate, decelerate, and steer the front wheel to maintain its balance much like humans do when learning to ride.
It’s captivating to see how the UMV uses its acquired skills. When it jumps, the robot's mechanical arm swiftly elevates the heavy control unit as high as possible, boosting the entire bike's momentum upward, and then that same arm pulls the bike into the air. For dismounting, it rides off the edge of a table and, mid-air, extends the arm again to lessen downward momentum, ensuring a gentle landing. The speed and fluidity of the movement are as impressive as the trick itself.
After executing the bunny hop, the robot performs a track stand, coming to a sudden halt and balancing motionless, just showing off. Then it does a cute little victory wheelie.
I'm uncertain about the practical applications for this technology, but there are potential real-world uses for this little machine. It could function as a delivery vehicle in busy urban settings or assist on job sites by carrying heavy tools and materials. However, the concept of machines learning to stretch the boundaries of their capabilities can be unsettling, especially if you’ve ever watched a sci-fi film. Let's just hope it doesn't evolve into a machine-gun-toting death robot in fifty years.
Got any tips? Send them to [email protected]


Other articles






Researchers Developed a BMX Robot That Learned to Bunny Hop on Its Own.
This device lacks gyroscopes or motion sensors; it learned to move, jump, and track its position through experimentation and experience.