USC researchers did present new humanoid robotics work in 2026, but the headline was a control model, not a new robot body, and the venue was an academic conference rather than a "summit." At Robotics: Science and Systems (RSS) 2026, USC Viterbi presented three peer-reviewed papers, led by Ψ0 (Psi-Zero), described by its authors as an open foundation model for universal humanoid loco-manipulation. "Loco-manipulation" simply means moving (locomotion) and handling objects (manipulation) at the same time — walking to a shelf and picking something up, for example. RSS is a selective, peer-reviewed conference, so acceptance signals scientific vetting, though not independent replication of every reported number.
Table of Contents
- What USC actually unveiled at RSS 2026
- Was it a "summit," and where was it held?
- How Ψ0 works and why the approach is notable
- How to read the performance claims
- What a practitioner can do next
- Frequently Asked Questions
What USC actually unveiled at RSS 2026
USC Viterbi presented three papers spanning humanoid loco-manipulation, robot foundation models, and 3D manipulation, according to USC Viterbi's July 2026 announcement. The standout is Ψ0, a software model that decides how a humanoid should move and act, not a new mechanical chassis.
This distinction matters for a robotics audience. USC's humanoid research develops Vision-Language-Action systems — software that turns camera input and language instructions into robot actions — and tests them on third-party robots such as Unitree and Dexmate platforms. In short, USC contributed the "brain," while the bodies come from outside vendors.
Was it a "summit," and where was it held?
Not a summit. RSS 2026 was the 22nd edition of an academic conference, held July 13–17, 2026 at ICC Sydney and the University of Technology Sydney, per the RSS program overview.
Readers should also avoid confusing this with a different event. The IEEE-RAS 25th International Conference on Humanoid robots (Humanoids 2026) runs December 6–9, 2026 in Silicon Valley — a separate, humanoid-focused venue that USC did not present at in July. If you are tracking humanoid announcements this year, these are two distinct conferences on different continents.
How Ψ0 works and why the approach is notable
Ψ0 credit belongs to more than USC alone. It was built by the USC Physical Superintelligence (PSI) Lab together with NVIDIA and WorldEngine, so "USC researchers" tells only part of the story, as the project's GitHub repository documents. The training method is the interesting part. Ψ0 learns general movement from more than 800 hours of human video, then refines those skills on roughly 30 hours of How to read the performance claims
The reported gains are large. On the Ψ0 project page, the team claims over 40% higher success rates on complex tasks while using 10 times less training data than the prior state of the art. Treat these as self-reported benchmarks, not settled facts. Before repeating the numbers, keep these limits in mind:. If you work in robotics or automation, Ψ0 is open, so you can inspect it directly rather than rely on summaries. Start with the Ψ0 project page, paper, and code, which links the technical report and repository. From there, a reasonable evaluation path is to check the hardware requirements, review the training data pipeline, and test whether the 80-trajectory skill-learning claim holds on a task you care about. Because the model targets third-party platforms like Unitree and Dexmate, confirm compatibility with your own robot before committing engineering time. No. USC contributed Ψ0, a control and learning model, and tests it on third-party robots such as Unitree and Dexmate rather than a robot of its own. Yes. It is released as an open foundation model, with the paper and code linked from its project page and GitHub repository. No. They are self-reported, lab-benchmarked results tied to specific tasks and hardware, vetted through RSS peer review but not independently reproduced.What a practitioner can do next
Frequently Asked Questions
Did USC build a new humanoid robot?
Is Ψ0 available to use?
Were the 40% and 10× figures independently verified?
You Might Also Like
