OpenArm: A $6,500 Open-Source 7DOF Humanoid Arm — and Laundry Folding as the Community's De facto Benchmark
enactic's OpenArm (183 points) is a fully open-source, human-scale 7DOF robot arm — hardware under CERN-OHL-S-2.0 (full CAD: STL/STEP/Fusion 360), software Apache-2.0 across CAN control, ROS2, teleop, Isaac Lab and MuJoCo — at $6,500 for a complete bimanual system. Two design choices stand out: high backdrivability and compliance for safe human-robot contact, and the OpenArm Cell, "a standardized environment with unified background, lighting, and camera placement. Research performed using OpenArm can be reproduced around the world in consistent evaluation conditions" — in effect, they're trying to do for physical-AI benchmarks what standardized eval harnesses did for LLMs.
The reality check came early:
"$6,500 for OpenArm v2. Incredibly smooth movement, 4kg of load per arm. Its unclear how far the level of autonomous control has come so far. The video shows mostly tele-operated operation." — donpdonp
The laundry debate
The comment section fixated on laundry folding, and the split is a genuine disagreement about what robotics milestones mean:
"It's a huge milestone for robotic capabilities more generally... It requires a gentle-but-firm grip and extreme dexterity across a wide range of shapes, sizes, weights, densities and material types. If a somewhat affordable robot can fold some of your laundry in a reasonable amount of time, all of a sudden there are a lot more new and useful things it can do as well. It's a metric." — slfnflctd
"To me it does not seem like progress that, on a stressed and finite planet, humanity might "need" an energy- and resource-intensive machine to complete a task as physically undemanding (and even frivolous) as folding a sheet. But here we are." — bluebarbet
"I genuinely don't understand this binary thinking. You could easily turn it around: if we're cutting down on inefficiencies, laundry folding robots will give people more time to exercise." — brookst
The pro-benchmark side has the stronger technical case: deformable-object manipulation is a proxy for the whole contact-rich manipulation problem, and showerst supplies the demand-side reality ("Laundry folding is probably family's most disliked trivial chore"). The deeper robotics critique came from moffkalast, who argues the field still hasn't solved base_link-to-gripper calibration or sag compensation even on the popular SO-ARM101 — "VLAs just use inverse kinematics as a crutch and predict gripper offsets, so we're back to square one."
Why it matters
Open hardware at $6,500/bimanual with reproducible eval cells is a real input-price collapse for physical-AI research — the same democratization dynamic HN has tracked in LLMs, now in contact-rich manipulation (compare The RISC-V "99%" Reality and the Hacker News Demographics Crisis for the ISA version of this argument). Watch whether OpenArm Cells actually produce comparable cross-lab results; if they do, the laundry benchmark — "wouldn't the companies just laundryfold-maxx their robots then?" (prometheus1992) — becomes Goodhart's-law territory the robotics community hasn't yet navigated.