Google develops ping-pong robot, which can catch the ball 340 times in one round.

CTOnews.com, Oct. 20 (Xinhua)-- yesterday, Google announced that it is working on a "ping-pong robot" project that can catch 340 times in a round against humans.

Currently, Google insists that this is just a "partnership" between humans and AI, not defeating humans, but at the rate at which AI is growing, it will soon target professionals.

The project, called i-Sim2Real, is not just about table tennis, but about building a robotic system that can work with fast-paced and relatively unpredictable human behavior. The advantage of table tennis is that it is severely restricted (compared to playing basketball or cricket) and strike a balance between complexity and simplicity. In this process, machine learning models are taught what to do in virtual environments or simulations, and then apply this knowledge, with the goal of catching the ball with humans for as long as possible without making mistakes.

CTOnews.com learned that in addition to i-Sim2Real, a method of alternating simulation and reality, Google researchers are also exploring a method of learning using only real-world data, the GoalsEye project, in which the former can alternately learn strategies between simulation and reality, while the latter learns from real-world unstructured data, combined with self-training, is effective for learning target conditional strategies in precise and dynamic requirements.