Frequently Asked Questions

Register a team of 1–6 on the registration page. Member 1 is the point of contact and receives all team email.

No — participation is free.

Teams are 1–6 people; solo is allowed, and you form your own team. Looking for teammates? Use our Discord (and the Shanghai testbed WeChat group for participants in China).

All levels are welcome — we provide baseline examples and tutorials.

Yes — submit the registration form again with all your details, including the corrected ones. Updates are accepted only through a complete re-submission (we use your latest submission), so please re-enter your full team info rather than sending changes another way. The team name must use Latin (English) letters and match across forms.

Your point of contact (Member 1) receives competition announcements and updates by email — there's no instant automated reply, so please allow some time and check your spam folder. If you registered and haven't heard from us, you're still registered; announcements go to every point-of-contact address.

Three independent tracks — Task 1: Cable Routing & Plugging; Task 2: Deformable Material Handling (Thermal Pad Placement); Task 3: Assisted Living & Feeding. Enter one or more; each is ranked independently.

NVIDIA Isaac Sim and MuJoCo. Not every task/simulator combination is ready in the developer preview — check STATUS.md for the current readiness matrix. (For example, the MuJoCo version of Task 2 isn't in the release yet; if you build your own, it still counts.)

Franka Duo GELLO, Keyboard Control, Foot Pedal Control, and VR Control.

Phase I — the open simulation benchmark — is fully remote; no travel. Phase II is real-world validation on Mobile FR3 Duo robots at partner labs (Hamburg, Munich, Pittsburgh, Shanghai): teams advancing from simulation get a hands-on bench-testing window (Aug 10–19, 2026), then from Aug 20–31 you keep submitting code while the organizers run the robots and evaluation.

The benchmark is available now as a developer preview — framework, simulation environments, robot assets, and task environments — at the benchmark repository. Check STATUS.md for which task/simulator combinations are usable today before you build. Still coming, released incrementally: the Teleoperation Toolkit, baseline models, tutorials and documentation, and real-world datasets.

Submissions are open now: submit your team's public repository (with a Dockerfile and a run README) through the GitHub issue form — full requirements are in the Submission section of the Competition page. Official scoring is run by the organizers following the rules on the competition page; evaluation code in the benchmark repository is a development aid, not the official scorer.

Three grant programs, open now on the compute grants page — Google Cloud (up to US$1,000 of Nvidia GPU credit), Computational Freedom (RTX 4090 / L40S, mainland-China teams), and AMD Instinct MI300X (shared 8× MI300X in Taipei). Mainland teams should use Computational Freedom or MI300X.

Not necessarily — the grants above are intended to cover compute needs.

A per-task prize for the Top 3 teams that complete the full AI workflow — data processing, training, inference, and official evaluation — on the AMD ROCm ecosystem, verified via your Docker image and the official AMD evaluation environment. It's optional and doesn't affect the overall ranking.

Discord, the website, and the WeChat group for participants in China.