Google DeepMind’s new AI robot plays table tennis at ‘human level’
Google DeepMind has trained a robot that can play table tennis at human level. This robot can match the skill levels of an average human player. This achievement showcases significant advancements in robotics and artificial intelligence.
Key Points
DeepMind’s AI learned to play table tennis through a combination of simulated and real-world practice. This allowed the robot to refine its skills and adapt to different opponents.
Google DeepMind trained its robot using a two-step approach: First, simulations were used to refine the robot’s hitting skills. Subsequently, real-world experience allowed for further improvement and adaptation.
Robot equipped with a 3D-printed paddle won 13 out of 29 table tennis matches against human opponents.
Robot won 100% of matches against “beginner” and 55% against “intermediate” players. However, it lost every single time that it faced an “advanced” opponent.
The robot faced challenges in hitting balls that were extremely fast, out of its visual range; or very low to the table due to safety protocols, instructing it to avoid collisions that could damage its paddle.
Background
In 2017, OMRON, a Japanese electronics company, introduced FORPHEUS, a “robot table tennis tutor” that was considered the first of its kind in the world.
This robot displayed how the relations between people and robots are likely to progress in the future.
Now, Google DeepMind has achieved a significant milestone in robotics by creating a robot capable of playing table tennis at human level.
The project entailed the creation of an advanced robotic system designed to detect the trajectory of the ball, anticipate the opponent’s moves, and perform accurate returns instantaneously.
By integrating simulation with actual training, the robot acquired proficiency in a range of table tennis skills.
This achievement showcases the potential of AI to excel in physically demanding tasks requiring rapid decision-making and adaptation.
Aim of the Company
Achieving human-level performance in real-world tasks is a key objective for the robotics research community.
This new table tennis robot represents a significant step toward that goal because table tennis is one of the most popular arenas in robotics due to the requirements for accuracy, planning, and speed.
These advancements could impact various areas of robotics and human-robot interaction, like developing generalist robots capable of performing many useful tasks such as manufacturing, healthcare, and interacting safely with humans in the real world.
News Gist
Google DeepMind has created a robot capable of playing table tennis at human level. This achievement showcases significant advancements in AI and robotics. The robot, trained through a combination of simulation and real-world experience, can compete against human players and demonstrates the potential for AI in complex physical tasks.