The robot lands on its feet like a cat, adjusting its position in mid-air. (Image Credit: Journal of Astronautics)
Researchers at the Harbin Institute of Technology developed a cat-like robot that twists and lands on its feet while falling. Powered by reinforcement learning, the team trained the robot to make mid-air posture adjustments. The team believes this could change how we explore celestial bodies in space.
This research addressed the challenges robots face when jumping on small asteroids. For instance, small leg force imbalances may result in crashes, uncontrolled spinning, or the robot rebounding off the surface. Jumping robots are the answer to those problems, according to scientists. However, that also means more limitations need to be dealt with first. In this case, a single jump keeps the robot airborne for ten seconds or longer.
That can cause small leg force imbalances, leading to uncontrolled spinning or the robot bouncing off the surface. So, the Harbin Institute of Technology team used reinforcement learning to train their robots. It took seven hours to train the AI to adjust movements for balanced landings. It managed to change orientation, achieving better roll (side-to-side tilt), yaw (rotational angle), and pitch (forward-backward tilt) in just a few seconds.
The cat robot demonstrated its capabilities. Its AI took eight seconds to stabilize after it launched at a steep 140-degree forward tilt. Additionally, the robot turned 90 degrees mid-air to change its direction. The team also developed a microgravity simulation platform to ensure the system works. They used air bearings to replicate an asteroid's low-gravity environment so the robot could float close to a nearly frictionless surface. Even though it’s only capable of 2D motion, the AI system proved effective based on the simulation results.
The entire system runs on low computational power. Combining that with its lightweight design makes the robot ideal for deep space exploration. In the future, the cat-like robot could have various applications, including asteroid mining. Before reaching that goal, the team needs to continue improving the AI to make it more adaptable to different terrain and environments.
Have a story tip? Message me at: http://twitter.com/Cabe_Atwell
