New footage shows Google‘s ping pong robot pulling off an epic 340-shot rally with a human.
The advanced robotic arm is seen effortlessly sliding along its side of the ping pong table as it returns the human’s shots.
The monumental rally ends when the robot sends a weak shot into the net, more than four minutes after it started.
Google used real-world and simulation-based training scenarios to turn the robot into a competent ping pong player.
It currently plays ‘cooperatively’ with a human to get a good rally going, but according to Google its skills could be built up to match the level of any human.
The robot, called i-S2R, is seen effortlessly sliding along its side of the ping pong table as it returns the human’s shots
The robot uses a training platform called i-Sim2Real, short for ‘simulation to real life’.
Researchers had wanted to train it up in a simulation so it could play with humans in the real world, but they had a ‘chicken and egg’ problem.
‘How do we gather examples of a human interacting with a physical robot in order to model human behavior in simulation without already having a robot that can interact with a human?’ said Google engineer Laura Graesser.
Engineers therefore alternated between training in simulation and deploying it in the real world until it was good enough.
‘In each iteration, the human behavior model and the policy are refined,’ said Graesser. ‘We stop iterating when we can finally have fun playing with the robot.’
Google says future work would build a robot ‘coach’ that could adapt its play style according to the skill level of the human participant to keep things ‘challenging and exciting’.
However, engineers admit that table tennis is a ‘high-speed, dynamic task’ that requires the two players to react quickly to each other’s moves, making it a ‘challenging test bed for research on human-robot interaction’.
The monumental rally ends when the robot sends a weak shot into the net, more than four minutes after it started
In the video, the human engineer is generally sending shots down the centre of the table, although Google says it can ‘match the precision of amateur humans’ in its returns and placement.
Both man and machine are using a right-handed forehand orientation the whole time, but regardless the bot displays some impressive reaction speeds.
According to Google, i-S2R can also play table tennis opposite a human or another robot.
Google describes the technology as ‘a research platform for agile robotics’ that can answer questions about human-robot interaction and cooperation.
Google said it used real-world and simulation-based training scenarios to build up the robot’s skills
‘Robot learning has been applied to a wide range of challenging real world tasks, including dexterous manipulation, legged locomotion, and grasping,’ Google researchers say in a blog post.
‘It is less common to see robot learning applied to dynamic, high-acceleration tasks requiring tight-loop human-robot interactions, such as table tennis.
‘In this project, the goal for the robot is cooperative in nature – to carry out a rally with a human for as long as possible.’
This is not the first table tennis-playing bot; Japanese robotics company Omron has previously shown off its table tennis bot, called Forpheus, at CES in Las Vegas.
Forpheus has sensors that collect data about the opposing player’s facial expressions and physical movements to play a match.
Not all sporting ventures with robots have gone swimmingly, however; earlier this year, a chess-playing robot broke a child’s finger during a tournament in Russia.
