A robot capable of playing amateur level ping pong has been developed by a team of engineers at Google's DeepMind Project.

Using a combination of advancements in robot design, including artificial intelligence (AI), the researchers have created an AI-based ping-pong playing robot with what they suggest is the highest performance level ever for a robot.

To create the robot system, the team first used an industrial robotic arm dubbed the ABB IRB 1100, which can reportedly manipulate its arm and hand quickly, while also quickly sliding side-to-side on a rail, thereby making it an ideal ping-pong-playing candidate.

To enable the robot to play ping pong, the team used a two-tiered approach wherein the first tier involved executing table tennis moves, while the second tier focused on strategy.

According to developers, its brain architecture combined several skill level controllers to focus on each of the types of moves necessary to return a ping-pong ball correctly. Each controller was trained using AI routines that learned by watching physics simulations and humans playing the game. This approach reportedly enabled the use of a small data set. Further, the team also employed an iterative process to refine the robot's skills.

During trials, the robot played human players of variable skill sets, handily beating all the beginner-level humans but none of the players with a high level of skill. The ping-pong playing robot reportedly won just over half of its matches against intermediate players.

The ping-pong playing robot is detailed in the article, “Achieving Human Level Competitive Robot Table Tennis,” which appears in the journal arXiv.

For more information, watch the accompanying video that appears courtesy of Google’s DeepMind Project.