Nori Robotics is dropping a $1,688 humanoid robot for researchers

Building a robot that can actually do things usually requires a lab budget in the hundreds of thousands, which makes large-scale dataset collection nearly impossible for most people. Antonio from Nori Robotics (YC S26) is trying to break that bottleneck by releasing a bimanual mobile robot for just $1,688. This isn't just a toy; it’s a specialized development platform designed specifically for people working on imitation learning, vision-language-action (VLA) models, and general robotics research.

The hardware specs are surprisingly dense for the price point. Most low-cost bots skip the complexity, but Nori went for a high degree-of-freedom approach to make it useful for real-world manipulation tasks.

• Degrees of Freedom: 19 total
• Arms: Dual 7+1 DOF arms with 1.5 kg payload capacity each
• Mobility: Differential wheeled base (a conscious trade-off to keep costs down vs. legs)
• Lifting: 55 kg capacity telescoping lift
• Vision & Sensing: Four 720p/30fps RGB cameras and 2D lidar
• Compute: Raspberry Pi 5 (4 GB RAM) handling SLAM and safety protocols
• Audio: Dual microphone array with full-duplex voice support
• Power: 432 Wh battery

The engineering decisions here are a masterclass in cost-optimization. To hit that sub-$2,000 target, they avoided expensive Quasi-Direct Drive (QDD) motors in favor of high-ratio servos. They also swapped bipedal legs for a wheeled base, which significantly lowers the mechanical complexity and failure points. Since the onboard Pi 5 is mostly for low-level tasks, any heavy-duty AI workloads—like running complex ACT (Action Chunking with Transformers) or VLA models—need to be offloaded to a local computer via LAN or a remote server via WAN.

If you are looking to get started with a practical tutorial on robot control or want to test your own AI workflow on physical hardware, they have some solid resources available:

The SDK is open-source and written in Python, which makes it pretty accessible for anyone comfortable with standard robotics stacks.

# Link to the SDK repository
https://github.com/Nori-Robotics/nori-sdk-py

They also provide a browser-based simulator, which is a huge win if you want to test your code before risking a physical crash.

https://lab.norirobotics.com/nori/model

Even at this price, the hardware is already demonstrating utility in basic tasks like opening drawers, restocking shelves, and even pouring drinks. The long-term vision seems to be a decentralized learning network where users without deep robotics expertise can "teach" the robot tasks via demonstration and share those skills across the entire fleet.

For those interested in the deep dive of how they actually engineered the cost structure, they've published a hardware paper here:

https://doi.org/10.48550/arXiv.2605.16537