Workshop on Agentic Robotics / Demo Track

Demo Platforms

We bring the robot. You bring the agent.

Details for the demo session. This supplements the Call for Contributions — submission goes through OpenReview by September 30, 2026.

Bring your own

Your own robot or portable setup — always welcome, and independent of what we provide. The demo itself runs on hardware; simulation belongs in the submission, not on the floor.

Use a provided platform

Run your agent on hardware we bring to Austin. Two platforms, below.

Provided Platforms

Platform A / AgileX Robotics

Dual-arm AgileX NERO

Two 7-DoF NERO arms on a shared column — the default tabletop and bimanual station, with a parallel-jaw gripper or a Revo2 dexterous hand per arm. Open to every accepted demo; no prior experience required.

Start developing in simulation

Participant preview: build your code-as-policy or agent-as-policy demo with the shared NERO Python and MCP interfaces. Follow the Linux x86_64 quick start, then test against the simulator before coordinating hardware integration with the organizers.

Quick start →GitHub repository →Participant guide & API →

Two AgileX NERO 7-DoF arms on a shared torso column — Vendor render. Final station configuration is confirmed closer to the workshop.

Arms: 2 × NERO, 7 DoF, 3 kg / 580 mm each
End-effector: AgxGripper or Revo2 hand
Bus: CAN
Teleoperation: Optional leader–follower lead arm

Full specifications→

Platform B / Unitree Robotics

Unitree G1 humanoid

A humanoid for demos needing locomotion, whole-body motion, or reach beyond a table. We cannot teach the platform on the day, so this slot requires video of your system running on a physical G1.

Robot: Unitree G1 — configuration TBD
Hands: TBD
Comms: DDS over Ethernet
Eligibility: Prior physical-G1 video required

G1 developer documentation→

What to Bring

If you use our platform

We bring the robot, mounted and powered — for NERO, your choice of gripper or dexterous hand
Compute is yours. Bring a laptop or workstation and run your stack on it
Camera: use the robot’s onboard camera, or bring your own — we recommend your own, with a mount, if your method depends on the sensor or viewpoint
You propose the task and bring the props it needs
A table, mains power and a wired network drop at the station

If you bring your own hardware

Tell us your venue needs — floor or table space, power, network, anything unusual
Everything else is yours to plan, shipping included
Same submission route; no platform slot to compete for

Logistics

Demo session: 16:00 – 18:00, November 12, 2026
Setup window: TBD — updated here once the venue confirms
Slot length: TBD — depends on the number of accepted teams

FAQ

Which platforms, and which tasks?: The two platforms above, plus bring-your-own. There is no fixed task list — demos live within tabletop manipulation, mobile manipulation and navigation, and within that scope you propose the task and bring the props it needs. Tell us early if it needs anything from the venue.
Do we get raw camera data?: Yes — raw images from the robot’s onboard camera, with calibration. We recommend bringing your own camera and mount if your method depends on a specific sensor or viewpoint.
What does the API expose?: For NERO, the workshop stack provides Python functions and MCP tools for RGB-D capture, pixel deprojection, IK/FK, joint and TCP motion, independent gripper control, and operation feedback. Simulation and hardware share the same interfaces. Policies receive no simulator object coordinates or scoring state; perception and task planning remain part of your demo. G1 uses the official Unitree SDKs linked above.
Is there a simulator before the workshop?: Yes. The NERO participant preview includes a MuJoCo simulator with place, select, and sort tasks, plus Python and MCP policy examples. Start with the quick start on Linux x86_64. Generated Python execution additionally requires Linux Landlock ABI ≥3 and libseccomp. Physical integration is coordinated with the organizers; success in simulation does not guarantee success on hardware. G1 reference simulators are linked above.
What compute should we plan for?: Bring your own laptop or workstation and run whatever you like on it — your models, your stack, your environment. The station provides power and a wired drop.
Can the submission video be simulation-only?: Yes, with physical integration after acceptance — except for the provided Unitree G1, where physical-robot video is the eligibility bar.
Can the same project go to both the Poster and Demo tracks?: Yes.
Are the tracks archival?: No. Contributing here does not preclude submitting the full manuscript to a journal or another archival venue later.
What is the deadline?: September 30, 2026, 11:59 PM Anywhere on Earth, via OpenReview.

Still marked TBD and you need the answer? Ask — TBD — workshop contact address.

Submit a demo Back to the workshop