4-legged robots are becoming a familiar sight, but engineers are still working out all the capabilities of these machines, and now, a MIT research team he claims that one way to improve their functionality could be to use AI to teach robots how to walk and run.
Usually, when engineers create the software that controls the movement of 4-legged robots, they write a set of rules about how the machine should respond to certain inputs, so if a 4-legged robot’s sensors detect x amount of force on “Leg y”, it will respond by starting the motor a to exert torque b and so on.
Coding these parameters is complicated and time-consuming, but offers researchers a precise and predictable control on 4-legged robots. An alternative approach is to use machine learningspecifically a method known as reinforcement learning that works through trial and error.
This works by giving your AI model a goal known as a “reward function” (eg move as fast as you can) and then letting it figure out how to get that result from scratch. This is time-consuming, but it’s useful to let the AI experiment in a virtual environment where you can speed up time.
That’s why reinforcement learning, or RL, is a popular way to develop an AI that plays video games.
This is the technique used by MIT engineers, creating new software (known as a “controller”) for the university’s 4-legged research robot. Mini Cheetahwho using reinforcement learning, was able to achieve one new top speed of 3.9 m / sor about 8.7 miles per hour.
The evolution of MIT’s 4-legged robot, Mini Cheetah
The new racing gait of the 4-legged robot, Mini Cheetah, is a bit clumsy, in fact it looks like a puppy climbing to accelerate on a wooden floor but, according to the MIT doctoral student Gabriel Margolis (a co-author of the research along with postdoctoral fellow Ge Yang), this is because AI is optimizing nothing but speed.
“RL finds a way to run fast, but given an unspecified reward function, he has no reason to prefer a” natural-looking “or human-preferred gait.”
says Margolis, further states that the model could certainly be tasked with developing a more fluid form of locomotionbut the whole point of the effort is to optimize only for speed.
Margolis and Yang say that a big advantage of developing controller software using AI is that it takes less time than messing around with all of the physics.
“Programming how a robot should act in every possible situation is simply very difficult. The process is tedious because if a robot were to break down on a certain terrain, a human engineer would have to identify the cause of the failure and manually adapt the robot controller “
they claim.
Using a simulator, engineers can place the robot in any number of virtual environments, from solid flooring to slippery rubble, and let it sort itself out. In fact, the MIT team says its simulator was able to accelerate for 100 days of staggering, walking, and running in just three hours of real time.
Some companies developing robots with legs are already using these kinds of methods to design new controllers. Others, however, such as Boston Dynamics, apparently rely on more traditional approaches. (This makes sense given the company’s interest in developing very specific movements, such as the jumps, vaults, and flips seen in its choreographed videos.)
There are also faster legged robots out there. Boston Dynamics’ Cheetah bot currently holds the record for a quadruped, reaching a speed of 28.3 mph, faster than Usain Bolt. However, not only is the Cheetah a much larger and more powerful machine than MIT’s Mini Cheetah, but it achieved its record by running on a treadmill and mounted on a lever for stability. Without these perks, perhaps the AI would give the car a run for its money.
If you are attracted to science or technology, keep following us, so you don’t miss the latest news and news from all over the world!
#4legged #robots #turning #point