This robot can learn from mistakes like humans!

According to reports, our understanding of general robots, if they want to add new features, need to add new procedures, and now it may be passive learning such as watching videos or hand-handling. The BRETT of UC Berkeley allows robots to actively learn in Trial and Error using neural network-based deep learning algorithms.

For example, when assembling a toy, the robot will try again until it knows the principle of assembly. In theory, this robot does not need to rely on the update of the human hand, but gives enough time for it to learn.

Of course, this sci-fi-like bridge is not ready for official launch. It now takes 10 minutes to confirm the process, and then try to make mistakes in 3 hours.

After all, BRETT has no rich experience for a while, so it can't have a logical leap to speed up understanding. But researchers will collect more data and experience for robots in the next few years. Ultimately, artificial intelligence will allow robots to do their best in the range of functions they design.