Google develops ping pong robots that can play 340 times in a row!

October 20, Google announced that its "table tennis robot" project i-Sim2Real is being studied, and its robot can catch the ball 340 times a round when playing against humans.

It is understood that Google chose table tennis because the robot can interact with fast-paced and relatively unpredictable human behavior.

At the same time, robots require both speed and accuracy to complete table tennis actions, which places high demands on learning algorithms.

Sports like table tennis have a fixed, predictable environment, making them ideal testbeds for studying human-computer interaction and reinforcement learning problems.

In addition, Google's table tennis robot can continuously learn in a simulated environment and apply the learning to the real world.

i-Sim2Real is also not just about ping-pong robots, but a way to create processes for artificial intelligence.

In addition to i-Sim2Real, an approach that alternates between simulation and reality, researchers are also exploring ways to learn using only realistic data, namely the GoalsEye project.

The former allows alternating learning strategies in simulation and reality. The latter allows learning from reality combined with self-training to achieve more accurate goals.