Table Tennis Agent

Closed weights Google DeepMind 185K parameters August 2024

No estimate

No hardware requirements for this model

The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
Google DeepMind
Organisation type
Industry
Country
United States of America
Published
7 August 2024
Authors
David B. D'Ambrosio, Saminda Abeyruwan, Laura Graesser, Atil Iscen, Heni Ben Amor, Alex Bewley, Barney J. Reed, Krista Reymann, Leila Takayama, Yuval Tassa, Krzysztof Choromanski, Erwin Coumans, Deepali Jain, Navdeep Jaitly, Natasha Jaques, Satoshi Kataoka, Yuheng Kuang, Nevena Lazic, Reza Mahjourian, Sherry Moore, Kenneth Oslund, Anish Shankar, Vikas Sindhwani, Vincent Vanhoucke, Grace Vesom, Pen…

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Robotics
Task
Sports
Approach
Reinforcement learning

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
185K

17 low level controllers with 10k parameters each: "Each policy is a dilated-gated CNN [22] following the architecture in [23] with 10k parameters... The final system contained 17 LLCs" One high-level controller with 4.5k parameters: "The style policy architecture, similar to the LLC but with only 4.5k parameters, has a (8, 128) observation space" spin classifier that is a 2-layer MLP of hidden sizes (128, 64) and input size 18, which is 10k parameters per o1 and Claude. So ~185k parameters …

Training data
2,400,000,000 tokens

~18k ball states "This iterative cycle of training models in simulation on the latest dataset, evaluating it in the real world, and using the annotated evaluation data to extend the dataset, can be repeated as many times as needed. We completed 7 cycles for rally balls and 2 cycles for serving balls over the course of 3 months with over 50 different human opponents, leading to a final dataset size of 14.2k initial ball states for rallies and 3.4k for serves. A summary of the dataset evolution i…

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights
Closed — provider access only
Model access
Unreleased
Training code
Unreleased

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Why it is tracked
SOTA improvement

"first learned robot agent that reaches amateur human-level performance in competitive table tennis"

Record confidence
Likely

Sources

Where this record came from and when it was last checked.

Reference
Achieving Human Level Competitive Robot Table Tennis
Last updated
28 November 2025

What the numbers mean

Where it came from

Table Tennis Agent was published by Google DeepMind, in United States of America, in August 2024. industry is the category the publisher falls under.

It works in Robotics, and is recorded as doing sports.

Its weights were never published, so it can only be reached through its provider. No graphics card changes that.

Training and provenance

The training set ran to roughly 2,400,000,000 tokens.

It is tracked in the underlying dataset for one reason in particular: sOTA improvement.

Answers

Table Tennis Agent — common questions

01

What is Table Tennis Agent used for?

Table Tennis Agent works in Robotics, and is recorded as handling sports. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

02

What GPU do I need to run Table Tennis Agent?

None. Table Tennis Agent is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.

03

Is Table Tennis Agent open source?

No. Table Tennis Agent has not had its weights published, so it exists only as a service controlled by its owner.

04

How many parameters does Table Tennis Agent have?

Table Tennis Agent has 185K parameters. 17 low level controllers with 10k parameters each: "Each policy is a dilated-gated CNN [22] following the architecture in [23] with 10k parameters... The final system contained 17 LLCs" One high-level controller with 4.5k parameters: "The style policy architecture, similar to the LLC but with only 4.5k parameters, has a (8, 128) observation space" spin classifier that is a 2-layer MLP of hidden sizes (128, 64) and input size 18, which is 10k parameters per o1 and Claude. So ~185k parameters total. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

05

Who created Table Tennis Agent?

Table Tennis Agent was published by Google DeepMind, based in United States of America, categorised as industry.

06

When was Table Tennis Agent released?

Table Tennis Agent was published in August 2024. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

Source

Original publication

Record last updated 28 November 2025

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.