RobotForge

Open-source robotics, end to end

Showcase

The robots, grippers, and tools we ship working demos for. Click through to inspect the URDF, drive the joints, or chat the design into CAD — all in your browser, no install.

Featured · open hardware

SO-ARM101 + Robonine 76 mm parallel gripper

The LeRobot 5-DOF arm with a 3D-printable parallel-jaw gripper that grabs anything from screws to soda cans. 150 N grip, 76 mm full stroke, 0.1 mm repeatability, swappable RealSense / Orbbec / USB camera mount. ~$70 in parts, 30-minute assembly.

Grip force
150 N
Full stroke
76 mm
Repeatability
0.1 mm
DOF (gripper)
1
Mass (PLA, 30%)
250 g
Cost (BOM)
~$70

Hardware design © Robonine · GPL-3.0 · STL/STEP/URDF in their repo. We host their assembly STL for the live preview to the right.

⚙️

loading STL viewer

Robonine parallel gripper mounted on the SO-ARM101 robot arm
Gripper mounted on SO-ARM101 — photo by Robonine
Robonine gripper 3D dimension diagram: 128 × 109 × 130.5 mm, 76 mm stroke
128 × 109 × 130.5 mm · 76 mm stroke

Featured · AI-driven design

Text → CAD, in your browser

Type what you want, the model emits structured ops, the scene updates live. Inspired by earthtojake/text-to-cad and Adam-CAD: natural language goes in, validated CAD ops come out, exports drop straight to STL / STEP / URDF — round-trip into the viewer, the simulator, or your printer.

boolean · primitives

make a 30 mm cube with a 12 mm hole through the middle

parametric · brackets

design a Raspberry Pi mounting bracket — 4 holes on a 58 × 49 mm pattern

assembly · robot parts

build a chibi rover chassis with two wheels and a sensor mast

manipulator · gripper

add a 76 mm stroke gripper finger, 70 mm long, with a soft pad

Open Forge: Text → CAD →Manual scene editorAnthropic / OpenAI / Gemini — bring your own key, stored in your browser, never sent to us.

Featured · ecosystem

Round-trip into Hugging Face LeRobot

The PyTorch-of-physical-AI hub — open imitation learning, ACT, Diffusion Policy, π0/π0.5, SmolVLA, GR00T N1.5 — plus the dataset format we export from our simulator. Drop your captured trajectories straight onto the Hub and fine-tune a real policy. SO-ARM100/101 native support means our viewer demo and their hardware are the same robot.

GitHub stars
23.7k
huggingface/lerobot
SmolVLA pretrain
481 datasets
10.6 M frames
Hackathon 2025
3,000+
participants · 100+ sites

For builders shipping products

Yes, these are commercially viable. Here's the licensing matrix.

For your own commercial product: stick to Apache-2.0 / MIT components for the SHIPPED hardware and software. Use GPL-3 grippers (or design your own) for showcase / education. Get a lawyer before finalising the BOM if you're distributing physical kits.

Why these together?

The LeRobot SO-ARM is the cheapest credible 5-DOF arm you can build at home. The Robonine gripper turns it into a manipulator that can actually pick things up. Text-to-CAD gives you a way to design mounts and accessories without learning Onshape. And the LeRobot dataset format means every demo you record in our simulator can fine-tune a real policy on a real arm. RobotForge stitches the loop: design in CAD → inspect URDF in the viewer → record trajectories in the simulator → export to LeRobot → run on hardware.