VLA Gallery / All models / Humanoid VLA

Tesla Optimus

Tesla · 2024-05(2차)

기타Humanoid VLANot disclosedNot disclosed

No official technical disclosure; per secondary sources, an end-to-end neural network using 2D cameras and hand tactile·force sensors runs in real time on the FSD computer.

Paper

No paper

Overview

Category

Humanoid VLA

Type

기타

Released

2024-05(2차)

Organization

Tesla

Open weights

No

Parameters

Not disclosed

Architecture

Backbone LLM

Not disclosed

Vision encoder

—

Action head

Not disclosed (Not disclosed)

Chunk

Not disclosed

Control rate

Not disclosed

Architecture (original summary)

Not disclosed ('end-to-end neural network'·'single neural network', secondary)

Training

Pretraining accelerators

Not disclosed

Pretraining duration

Not disclosed

Pretraining steps & batch

Not disclosed

GPU/TPU-hours

Not disclosed

Fine-tuning

Not disclosed

Data type

Real-robot teleoperation (secondary) / human video (secondary)

Data scale

Not disclosed

Main datasets

Not disclosed

Tasks & evaluation

Battery-cell sorting (2024-05, secondary); household chores from egocentric human video (2025-05, secondary); no metrics

Robot

Tesla Optimus humanoid

Inference

Inference device & speed

Onboard FSD computer (secondary); speed not disclosed

Inference memory

Not disclosed

Weights bf16 (calc.)

Not disclosed

Where it runs

Onboard (FSD computer, secondary)

Reinforcement learning & base

Reinforcement learning

Not disclosed · No official technical documentation

Base control

Humanoid legged (Format not stated) · Mentions only the balance·navigation software stack of the biped humanoid; locomotion network architecture·command format not disclosed

Notes

Key features

No official technical disclosure; per secondary sources, an end-to-end neural network using 2D cameras and hand tactile·force sensors runs in real time on the FSD computer. In 2025-05 it said a substantial part of learning is transferred directly from human video (secondary).

Notes

Model not released. Release month based on the 2024-05 battery-cell sorting video (secondary-source date). The '48 networks, 70,000 GPU hours' on Tesla's AI page are FSD values and must not be used for Optimus. Type·architecture Not disclosed, so type is Other and architecture is Not disclosed. Not covered by verification files. Secondary sources: https://www.notateslaapp.com/news/2761/teslas-optimus-is-now-learning-by-watching-first-person-today-third-person-tomorrow-video ; https://www.digit.in/news/general/watch-tesla-optimus-robot-sorting-batteries-strolling-through-factory.html

Verification

Not verified