An international team of researchers presented a robot that can adapt to difficult surfaces and move on them. The device coped with this at least 70% of the time.

The researchers explained that many robot models have difficulty adapting to the environment, making it difficult for devices to navigate unusual surfaces with obstacles. They also fail to stop abruptly and avoid hard falls. But now scientists have created a new robot model that adapts in real-time to any terrain, changing gait to keep moving. The device also changes the method of movement if the robot collides with sand, rocks or stairs.

A team from Facebook AI, UC Berkeley, and Carnegie Mellon University called this phenomenon “rapid motor adaptation.” This is a property that is inherent in humans and animals – they can quickly, effectively and unconsciously change the way they walk, depending on the circumstances.

The system was fully trained in simulation mode, where the robot’s brain was taught to maximize forward movement with minimal waste of energy. A separate task of the device is to avoid falls, for this it observed around itself and reacted to the data that comes from artificial joints, accelerometers, and other physical data.

The system consists of two parts: the main, constantly working algorithm, which actually controls the gait of the robot, and in parallel, an adaptive algorithm that monitors changes in the internal readings of the robot. When significant changes are detected, it analyzes them and issues commands to the main model. From this point on, the robot only thinks about how to move forward in new conditions, effectively improvising a specialized gait.

The robot was able to walk through sand, mud, hiking trails, tall grass and heaps of mud without a single glitch in all tests. The robot descended the ladder in 70% of the tests and climbed the cement heap 80% of the time. The device did the same with a payload of 12 kilograms, which is 100% of his body weight.