Without Trails

Facebook’s AI researchers are teaching robots to walk like humans


NEW DELHI: Researchers at University of California, Berkeley and Carnegie Mellon University’s School of Computer Science, along with social media giant Facebook’s AI research team, have developed a new way for robots to walk better. The method allows robots to walk on different surfaces, providing more stability and possibly more functionality. How to have robots adapt when walking on different services is a problem that many computer scientists are working on.

In a blog, Facebook says that the solution — Rapid Motor Adaptation — is a “breakthrough” in artificial intelligence (AI). It allows robots to walk on sand, in mud, hiking trails, tall grass and over a dirt pile “without a single failure” during trials. “It successfully navigated a cement pile and a pile of pebbles in 80 percent of the trials, despite never seeing the unstable or sinking ground, obstructive vegetation, or steps during training,” the post said. “It also maintained its height with a high success rate when moving with a 12 kg payload, which amounted to 100 percent of its body weight,” it added.

To do this, the researchers used two branches of AI, called supervised learning and reinforcement learning. Supervised learning depends on providing algorithm data about situations it will face, which it can refer to in order to make decisions. Reinforcement learning includes actually having the AI interact with the environment and learn from its own decisions. Facebook says that the robot demonstrated “an aptitude fundamental to all intelligent agents”, which loosely means it was closer to emulating how humans adapt to terrain while walking.

The robot is able to determine the amount of friction a surface offers, the weight it’s carrying, abnormalities in the terrain, etc. and adapt its movement accordingly, and without actually looking at the surface. The second part is important because if successful, it eliminates the need to put cameras on such robots, thereby reducing the cost of building them.

According to Facebook, robots that need to walk on a particular terrain are specifically designed for such terrain, at the moment. For instance, robots working on the factory floors have algorithms that are adapted to that environment. With RMA, Facebook claims the robots can adapt to the environment “from scratch” by “exploring and interacting with the world”. It’s apparently the first such solution in the world.

Researchers at Norway’s University of Oslo had also showcased a similar solution in March, meant to enable robots to adapt to their environments. That solution included using machine learning (ML) and AI to change the robot’s leg length and its body shape to adapt to the terrain. The solution essentially aimed to adapt how humans change their centre of gravity dynamically while walking on different terrains. Like Facebook, the Oslo researchers also showcased four-legged dog-like robots, but their solution depended on 3D cameras and force sensors. Facebook didn’t disclose what kind of sensors its robots require.

Subscribe to Mint Newsletters

* Enter a valid email

* Thank you for subscribing to our newsletter.

Never miss a story! Stay connected and informed with Mint.
our App Now!!


Read More:Facebook’s AI researchers are teaching robots to walk like humans