The $1,688 Humanoid: Nori Robotics' YC-Backed Bid to Collapse Robot Economics

While the humanoid robotics headlines of 2026 have been dominated by five- and six-figure machines from Tesla, Figure, and Unitree, a five-person Y Combinator startup called Nori Robotics just pulled the price floor out from under the entire category. The company, part of YC’s Summer 2026 batch, is now shipping a bimanual mobile manipulator — it describes it as a humanoid, though it rolls rather than walks — for $1,688, roughly 3% of the cost of a typical capable humanoid, which the company notes starts above $20,000.

The launch, posted to Hacker News by founder Antonio Sitong Li, gathered 193 points and a lively 65-comment thread within a day. Behind the modest point count sits an immodest thesis: the thing blocking general-purpose robotics from homes isn’t intelligence anymore — it’s cost discipline.

The hardware: what $1,688 actually buys

After seven hardware iterations, the current Nori is a deliberately constrained machine. The full spec sheet from the launch post:

• 19 degrees of freedom total
• Two 7+1 DOF arms, each with a 1.5 kg payload
• A 55 kg telescoping lift for vertical reach
• A differential wheeled base (no legs, no balance controller)
• Four 720p 30 fps RGB cameras
• 2D lidar for navigation
• A dual microphone array with full-duplex speech dialogue
• A 432 Wh battery
• A Raspberry Pi 5 with 4 GB RAM as the onboard computer

That last spec is telling. SLAM and safety systems run on the Pi, but anything computationally heavy — action chunking transformers (ACT) or vision-language-action models (VLA) — has to run from a tethered computer over LAN or a server over WAN. Nori is not trying to be a brain; it’s trying to be an affordable body.

How they hit the price

Getting a bimanual mobile robot under $2,000 was, by Li’s own account, the central engineering problem. The robot has more than 100 moving and structural parts, and costs compound quickly across actuators, bearings, wiring, power delivery, and assembly. Two design decisions dominate the cost reduction:

First, high-ratio servos instead of quasi-direct-drive (QDD) motors. QDD actuators — the kind used in premium research humanoids — offer backdrivability, torque transparency, and smooth control, at a price that alone can exceed Nori’s entire bill of materials. RC-style servos are cheap but sacrifice force feedback and fine motion control, a trade-off the Hacker News commentariat seized on immediately.

Second, wheels instead of legs. Bipedal balance is one of the most expensive problems in robotics, both in compute and in actuator quality. A differential drive base eliminates it entirely. One commenter arguing for a home deployment put it bluntly: when a legged robot like the Unitree G1 falls, “the result is catastrophic” — the balance controller can lash out with dangerous force near children. A wheeled machine that physically cannot fall has a safety story legged robots can’t match.

The company assembles every unit in San Francisco and has designed the robot to be manufactured and repaired easily — it publishes 3D files so owners can print their own replacement parts, a deliberate nod to the right-to-repair crowd.

Open by default

Nori ships with an open SDK (nori-sdk-py on GitHub) covering teleoperation and demonstration collection, plus a browser-based simulator at lab.norirobotics.com so prospective buyers can try the platform before committing. The hardware itself is partially open source, and the company has published a hardware paper on arXiv (doi:10.48550/arXiv.2605.16537) detailing the design.

The openness isn’t charity — it’s the data strategy. Nori’s YC page is explicit about the endgame: “Nori aims to solve the robotics data bottleneck. By deploying affordable robots to customers around the world, we can gather the diverse data required to train the next generation of generalist robotic policies. Every user of Nori makes Nori smarter.” A fleet of thousands of cheap robots in real homes and businesses is a distributed data-collection engine that no single lab — hoarding one or two $100K robots — can replicate.

Traction and the market math

Six weeks after launching, Nori says it has deployed its first robot and booked over $350,000 in sales, with the next batch in production and shipping scheduled for fall 2026. The company’s market framing is aggressive but coherent: 132.7 million U.S. households plus 8.4 million employer business locations. Sell one $1,688 Nori to just 0.5% of them and you have a $1.2 billion business in the U.S. alone.

Founder Antonio Sitong Li studied computer science and architecture at Columbia, where he held research fellowships with the Laidlaw Scholars Program, the Data Science Institute, and the Graphics & UI Lab, and where his research focused on teaching robots tasks through VR demonstrations — the seed of Nori’s teach-by-demonstration software. He holds a national patent for a computer-vision system and previously co-founded Truely.

The skepticism, fairly stated

The Hacker News thread was not a victory lap. The sharpest technical criticism targeted the servo choice: no force feedback on positioning, jerky motion from actuator stepping, limited precision, and poor low-speed control — fine for coarse manipulation, inadequate for precise work. Others noted that basic servo control without proper trajectory generation is a safety liability on a machine this size, urging torque limiting and real e-stop behavior.

The demo videos also took fire. Commenters frame-by-framed the launch footage and complained that the folding, stirring, and restocking clips never actually show a task completed cleanly — the clothes get “awkwardly thrown into a pile,” the fridge door never closes. The company’s claim that the hardware is “already capable” of basic cleaning, drawer-opening, shelf-restocking, and beer-pouring was read by some as aspirational marketing wrapped around a development platform.

That reading isn’t necessarily fatal — Nori explicitly positions the product “for robotics developers and researchers,” and the price is low enough that buyers know they’re getting a starting point, not a finished butler. But it’s a reminder that the gap between a $1,688 body and a useful household robot is filled entirely with software that mostly doesn’t exist yet.

Why it matters

Nori’s real product is a bet on how generalist robotics gets built. The incumbent path is vertical: expensive humanoids, pilot deployments, data collected slowly at enterprise sites. The Nori path is horizontal: make the body absurdly cheap, sell it to everyone from hobbyists to hotel operators, and let the fleet generate the training data. It’s the PC argument applied to robotics — don’t sell capability, sell potential, and let a thousand developers build the capability for you.

The household-labor market the company cites is genuinely one of the largest unpriced markets in the world, currently served by a stack of single-task appliances — dishwasher, washer, vacuum — each costing $600 to $1,000 and doing exactly one thing. If Nori’s fleet model works, the sub-$2,000 bimanual robot could do to that stack what the smartphone did to the point-and-shoot camera, the GPS unit, and the MP3 player simultaneously.

And if it doesn’t work — if the servos are too jerky, the demos too optimistic, the data too noisy — the company will have still published its hardware paper, open-sourced its SDK, and pushed the cheapest credible bimanual platform ever shipped into the hands of exactly the people most likely to improve it. For $1,688, that’s a cheaper bet on the future of robotics than almost anything else on the market.