HumanoidGPT · Part of The Humanoid Group

Robot Foundation Models

Foundation models changed how software handles language and images. Robotics is applying the same idea: train one large model on broad data, then adapt it to many robots and tasks, instead of programming each skill by hand.

By Arjun Rao · Updated October 2026

What a foundation model is

A foundation model is a large neural network trained on a broad dataset so that it learns general patterns, which can then be adapted to specific jobs with far less additional data. In language, that means one model can summarise, translate and answer questions. In robotics, the hope is a model that understands scenes, instructions and physical actions well enough to transfer between tasks, environments and even different robot bodies. The term covers many approaches, so it helps to ask what a given model was trained on and what it actually controls.

Why robotics wants them

Traditional robot programming works well for repeated, tightly defined tasks, but every new object, layout or job can mean new code and careful tuning. Learned models promise more general behaviour: recognising unfamiliar items, recovering from small mistakes and following instructions in everyday language. For humanoids, which are meant to work in spaces built for people, that generality is the point. A robot limited to one scripted job is hard to justify in a general-purpose body.

What is still hard

Robot foundation models are an active research area, and real-world reliability remains the central challenge. Physical tasks have long tails of unusual situations, mistakes have real consequences, and there is far less robot data than text or images. Models also have to run fast enough to control a moving machine, often on hardware inside the robot. Expect steady progress rather than a single breakthrough, and judge any system by repeatable results on tasks like yours rather than by demonstration videos.

Sources and further reading

On the Opportunities and Risks of Foundation Models — Stanford Center for Research on Foundation Models (arXiv).
The Stanford report that introduced the term foundation model, covering capabilities, training methods and societal risks.
Foundation Models in Robotics: Applications, Challenges, and the Future — Stanford University, Princeton University and partners (arXiv).
Survey of how pretrained models feed robot perception, planning and control, and the gaps in robot data, safety and speed.
RT-1: Robotics Transformer for Real-World Control at Scale — Google Research (arXiv).
An early test of one transformer controlling real robots across many tasks, with results improving as data and model size grew.

Common questions

Is a robot foundation model the same as a chatbot?

No. It may share ideas and components with language models, but its output is ultimately robot behaviour: where to move, what to grasp and when to stop.

Does every humanoid use a foundation model?

No. Many robots combine learned components with conventional control and planning. Ask each maker which parts of the system are learned and which are programmed.

Robot Foundation Models — Large models trained on broad data, then adapted to control robots across many tasks.
Vision-Language-Action Models — Models that see a scene, read an instruction and output the robot's next movements.
Data & Simulation — Teleoperation, demonstrations and simulated worlds: where robot training data comes from.
Evaluating Humanoid AI — How to judge a humanoid's intelligence on your own tasks, not in a demo video.
LLMs as Task Planners — How large language models break a spoken request into steps a robot can actually carry out.
World Models for Robots — Learned simulators and video prediction that let robots imagine the result of an action before they act.
On-Device or Cloud AI — Where a robot's AI runs, and how latency, privacy, connectivity and power shape that choice.
Open-Source Robot Models — Open robot foundation models and datasets: what gets released, what open really means and how to use them.
AI Regulation for Robots — How the EU AI Act, NIST's AI Risk Management Framework and India's AI guidelines apply to robots.
Latest updates — Robot models are now tested in unseen homes and judged on safety, yet benchmarks still show wide gaps. These figures and dated updates show what progressed, what failed and how to read the claims.
About us — What we cover, how we source every figure, and who publishes this site.

Visit The Humanoid Group

People also search for robot foundation model, what are robot foundation models, foundation models for robotics, humanoid robot foundation model, AI model for robots and LLM for robots.