Google Robots Teach Themselves to Get A Grip

There’s been a lot of focus on machine learning in recent years. Machine learning is a type of artificial intelligence. It enables robots, computers, and automated systems to gather information, and use that information to perform new tasks or make decisions without explicit programming. This concept makes sense to us because it mirrors the way that people learn. We observe through teaching and experience, and we learn how to do new things. It’s difficult to simulate this type of learning in robots, however. With all of the time and effort researchers sink into machine learning, we have yet to see a robot capable of learning in a way that’s truly useful. But extra time and effort now could save time and effort in the future. Machine learning could eliminate the need for specific programming, and it could make robots much more versatile and capable than they are today.

Google and machine learning

We all know the work that Google does to improve machine learning. The world’s most popular search engine is machine learning in action, after all. A perpetually improving search engine isn’t the only way they contribute, however. Google’s research in robotics is also improving machine learning. Google connected 14 robotic arms together in a neural network to teach each other how to pick up objects.

Picking up items is a simple task for people, but it can be surprisingly difficult for robots. This is especially true when the objects vary in size, weight, or texture, and they aren’t in the same exact position every time. As the video above demonstrates, machine learning could help make this task easier for robots. The robots used information from successful attempts to improve their ability to grasp items. After 80,000 attempts, the robots started to correct their errors, and got better at their assigned tasks. The robots improved their success rate by 18% form learning through sharing and learning from each others mistakes.

Machine learning could benefit manufacturing

Google hopes to apply this information to real-world environments. Manufacturing would clearly benefit from this research. Clearly, these robots couldn’t cut it in the manufacturing industry today. Manufacturing requires speed and precision, and these robots are far less capable than, say, human workers would be. Once machine learning improves, however, this method could drastically improve the manufacturing process. Are you ready for Industry 4.0? The next chapter in manufacturing emphasizes automation and will be driven by data. Make sure your system can keep up. Call us today for any Indramat service or repair needs!