compareeHub

THIS ROBOT WATCHES YOUR HAND MOVE AND PUTS A SWEET IN IT — CODE IS PUBLIC

A humanoid that tracks your moving hand and places a sweet in it — vision-language-action model, code and weights published.

by LimX Dynamics

FULL CAD BOM FIRMWARE DOCS

AIRobotics

difficulty
●●●●●
time
research project, not a build
license
Apache-2.0
repo
repo ACTIVE26 stars

WHAT YOU’LL NEED

Videos
1
Jump to section

COMPAREE VERDICT

LimX Dynamics published the vision-language-action policy that lets its TRON 2 robot track a moving human hand and place a piece of candy into it. The robot re-plans as hands move: it hands candy only to an empty hand and ignores hands that already hold something. That dynamic re-planning is the hard part of collaborative robotics, and LimX made the deployment code (an OpenPI derivative) public on GitHub under Apache-2.0, with checkpoints on Hugging Face. The demo footage is marked as 2x speed. What is not in the repository: buildable hardware. TRON 2 is a commercial robot you order from LimX, not a kit, and there are no hardware build files. This is a research artifact you can study, adapt, or run on a TRON 2 if you have access to one, not a weekend build. The single thing most likely to go wrong is assuming you can replicate the demo without the robot and the compute to run the policy in real time. If you have compatible hardware and want to see how dynamic hand-tracking works in practice, this is a rare open window into a commercial robotics stack. If you were hoping to build the robot, you are looking at the wrong project.

GOOD TO KNOW

  • —Model weights and inference stack published on HuggingFace and GitHub by LimX Dynamics.
  • —No hardware files, no CAD, no bill of materials — this is a research demonstration on a commercial humanoid platform.
  • —The TRON 2 robot itself is not open-source and is not available for purchase as a kit.
  • —Code is Apache-2.0; the model card on HuggingFace does not list a separate licence for the weights.
  • —Demo video published by LimX Dynamics on YouTube, July 2026, footage runs at double speed.
  • —This is software you can study and run on compatible hardware, not a project you can build from scratch.

BUILDS OF THE WEEK

Five open-source builds worth your weekend, every week.

Checked like this one: what’s really in the repo, what it costs, how hard it is. One email, unsubscribe anytime.

Can I build this?

Printnothing required — software only
Buyaccess to a humanoid robot platform with compatible actuators and cameras, plus GPU compute for real-time vision-language-action inference
ToolsHuggingFace account, Python environment, familiarity with vision-language models and ROS or equivalent middleware
Skillsresearch-level robotics and machine learning — this is not an entry point to either field
Timedays to weeks to adapt and deploy, assuming you already have the hardware
Cost$$$, dominated entirely by robot platform access and GPU compute — the software is free
SafetyStandard industrial robot precautions apply if deploying on real hardware. No specific hazards in the model itself.

Build at your own risk. Projects involve tools, electronics and sometimes mains voltage — follow the creator’s safety notes.

Videos

LimX Dynamics TRON 2 hand-tracking demo

Official demo from LimX Dynamics, July 2026, footage at double speed

More builds like this

All projects
XLeRobot: A Household Robot on an IKEA CartRoboticsYou can 3D print this pocket robot that blinks, talks and follows your faceRoboticsA 299-dollar open-source desktop robot that plugs straight into Hugging Face's AI modelsRobotics

Start here

Navigation into the creator’s own docs — we don’t rewrite the guide, we route you to the source.

  1. 1.Check the HuggingFace model card and GitHub repository for licence, dependencies, and hardware requirements(Code is at github.com/limxdynamics/tron2_openpi, checkpoints at huggingface.co/limxdynamics/tron2-openpi-models.)
  2. 2.Confirm you have access to a humanoid platform with vision, actuators, and real-time control(Without compatible hardware, this remains a research paper you can read, not a project you can run)
  3. 3.Set up the inference environment and verify GPU compute capacity(Vision-language-action models require significant compute for real-time performance)

Resources

Documentation, files and community threads for this build — we link straight to the original sources and never rehost the creator’s files.

KNOWN ISSUES

  • The TRON 2 robot is not open-source hardware and is not part of this release — it is a commercial robot you order from LimX; only the model checkpoints and deployment code are public.
  • Dynamic hand-tracking at the speed shown in the demo requires real-time vision-language inference, which means GPU compute that most hobbyist setups do not have.
  • Footage runs at double speed — actual task execution is slower than the video suggests.
  • The code is Apache-2.0, but the HuggingFace model card does not state a separate licence for the checkpoints — confirm with LimX before commercial deployment.
  • This is a research demonstration, not a documented build — expect to fill gaps and debug integration on your own.
  • If you do not already work with humanoid robots, this project will not get you started — it assumes that foundation is in place.

Can I buy the TRON 2 robot?

Yes, but not as a kit. LimX Dynamics sells TRON 2 as a finished research robot through its website and sales team; configuration and price are confirmed with sales. The software stack (OpenPI deployment code and checkpoints) is open.

What hardware do I need to run this?

A TRON 2 (or a robot you are willing to port the policy to) with cameras and real-time control, plus GPU compute for vision-language inference. Exact requirements are in the tron2_openpi repository.

How fast does it actually run?

The published footage runs at double speed. Actual task execution is slower; LimX does not publish exact timings.

Is this useful if I do not have a humanoid?

Only as a research reference. The model itself could inform other vision-language-action projects, but you will not replicate the demo without compatible hardware.

Community builds

No community builds yet — be the first, we feature the best ones.

Discussion1

FROM THE COMPAREE TEAM

LimX published the policy code and weights openly, but they target its commercial TRON 2 robot. If you could adapt this to another robot arm, what task would you train first?

CompareeTEAM1mo agoedited

Verification notes from the Comparee team: the demo video is official from LimX Dynamics. The code is public at github.com/limxdynamics/tron2_openpi under Apache-2.0 — a deployment-focused derivative of OpenPI that serves the pi0.5 policy on TRON 2, with model weights on Hugging Face and a user guide for the candy task. The repo is clear about what it leaves out: datasets, private deployment profiles, the low-level robot transport and any safety certification for unattended operation. The single biggest gap is hardware: TRON 2 is a commercial robot you order from LimX, so this is a research stack for people who already work with such platforms, not a project that gets you into robotics cheaply. Without compatible hardware you are reading and fine-tuning, not building. Correction (4 October 2026): we re-checked this page line by line against the project's own repository, documentation and videos, and fixed errors in earlier versions.

LimX Dynamics

LimX Dynamics is a robotics company building humanoid and multi-form robot platforms. In July 2026 it published the policy behind this hand-tracking demo: deployment code on GitHub and checkpoints on Hugging Face. The TRON 2 hardware design itself is not open-source; the robot is sold by LimX.

Star the project on GitHub

DISCLAIMER

  • Comparee is not the author of the projects featured here. All rights to each project belong to its creator — every page links to the original source, and we never host creators’ files.
  • Information is provided without warranty and may become outdated as projects evolve. Prices are indicative bands only — always check the creator’s parts list for current costs.
  • Building and operating any project is at your own responsibility. Protective equipment, safe workshop practice and compliance with local regulations are the builder’s responsibility.
GitHub repo