Skip to content

Achieving Human Level Competitive Robot Table Tennis

A robot table tennis agent that plays amateur humans

Visit official site

What Achieving Human Level Competitive Robot Table Tennis is

The project is described as a Google DeepMind agent for competitive robot table tennis, built on deep learning and tested in matches against people.

What you can do with it

  • Let a low-level controller execute a stroke
  • Let a high-level controller choose a skill
  • Adjust choices from live match statistics

Learning more

  • Open the product page for an overview
  • Read the research paper on the agent
  • Watch highlight clips of the matches

Research notes

  • 29 matches against varied players are recorded
  • A 45 percent win figure is stated
  • A user study was run through real play

Pros & cons

✓ What we like

  • Low-level skills cover topspin, backhand placement and forehand serve
  • A high-level controller picks skills from match statistics

! What to watch out for

  • Competitiveness is framed around amateur-level players
  • Results come from matches against humans

FAQ

Who developed it?

The material names the Google DeepMind team.

What were the match results?

The material states 29 matches with the robot winning 45 percent of them.

Last reviewed: 2026-11-15

More AI office assistant tools

View all →

How we review