Atelier

We’re a small group of engineers and roboticists exploring ideas that are newly within reach. Most of us are Stanford researchers who also work in labs like ILIAD, REAL, and Hazy Research.

We’re starting now because the economics and the possibilities of robotics research are shifting.

1. Before an actual experiment, it used to take weeks to integrate components from several vendors, calibrate poses, and fabricate a test rig, to name a few. Frontier AI has collapsed the cost of software, and is starting to do the same for parts of hardware.¹ Most of the setup is no longer the bottleneck.
2. Frontier LLMs are also demonstrating resounding spatial reasoning capabilities, strong enough to generate realistic 3D environments and even actuate robots directly.² We think much of the field still underestimates how much value these models can unlock when used creatively.

Together, these mean that testing a strange idea is far cheaper than it was even a few months ago, and more ambitious research directions are now within reach.

The name comes from nineteenth-century Paris, where ateliers were small workshops for artists and architects to learn and create with one another outside of the academy. Much of the era’s most important work came out of independent ateliers like the Académie Suisse, where Monet and Cézanne trained.

We think robotics needs the same thing right now, and we envision Atelier as a contemporary equivalent for engineers and roboticists. The point is not to commercialize our work, but to provide a space with resources and other people working seriously on hard problems.

Early directions

Right now, we’re most excited about three directions. The first is unexplored end effectors and dexterous tool use. The field has mostly converged on parallel-jaw grippers and dexterous hands,³ and we think the space of possible bodies, and how one collects data for them, is much larger than what’s been tried.

The second is automating the novelty pump. Robot foundation models only improve with truly novel data, and today, finding the states and tasks a policy hasn’t seen is human operational work that doesn’t compound. Frontier models already know a lot about the world, and we think they can accelerate this process across the robotics and RL stack. We’re also interested in improving them as policies in their own right.

The third is what we call embodied alignment: making sure robots act in ways that follow our intent, including the parts we don’t think to specify.⁴

Reach out

If you have a bet you want to test and Atelier sounds appealing to you, we’d like to hear from you. You don’t need to be at Stanford or in a lab. Email us at mmusic@stanford.edu.

If you’re a lab or company interested in sponsoring research or learning more about what we’re up to, email us at mmusic@stanford.edu.