UnderControl AI
Last reviewed
May 11, 2026
Sources
10 citations
Review status
Source-backed
Revision
v3 ยท 2,433 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 11, 2026
Sources
10 citations
Review status
Source-backed
Revision
v3 ยท 2,433 words
Add missing citations, update stale details, or suggest a clearer explanation.
Noble Machines (formerly Under Control Robotics, also known as UnderControl AI or UCR) is an American robotics company based in Sunnyvale, California, United States, developing AI-driven humanoid robots for heavy industrial applications. Founded in May 2024 by former engineers from Apple, SpaceX, NASA, and Caltech, the company emerged from stealth on March 3, 2026, after deploying its first humanoid robots at a Fortune Global 500 customer within 18 months of launch. The company's flagship product is Moby, a bipedal humanoid robot designed for manufacturing, logistics, construction, energy, mining, and semiconductor industries. UnderControl AI built its technology stack on the NVIDIA Isaac platform and uses Arm architecture processors throughout its hardware.[1][2][3]
Under Control Robotics was incorporated in May 2024 by a small founding team of engineers who had worked at Apple, SpaceX, NASA, and Caltech. The four founders established the company after Wei Ding and Wenlong Ma returned to the United States from a period of work in the Shenzhen manufacturing hub, where Ding had been building robot manipulation systems for the textile industry.[3][4]
The company initially operated from a small 80 square meter lab in Santa Clara, California. Within roughly four months, the founding team had built its first functional bipedal robot, which they named Moby after the white whale in Herman Melville's novel. The robot was showcased publicly for the first time at the Bay Area Humanoid Summit in December 2024.[3][5]
During its stealth period, the company operated under the Under Control Robotics name (UCR) and worked from the small Santa Clara lab while iterating on Moby's mechanical platform and AI training pipeline. The Humanoid Hub, a robotics industry publication on X/Twitter, visited the lab in early 2025 and described meeting the four co-founders working alongside Moby in a single room.[5]
The team integrated arms onto Moby for object manipulation by late March 2025 and began testing the robot in scenarios involving steep inclines, scaffolding, and uneven outdoor terrain. The company invested heavily in sim-to-real transfer using the NVIDIA Isaac platform, achieving what it later claimed to be a 95% sim-to-real deployment success rate.[1][3]
On March 3, 2026, the company emerged from stealth under a new name, Noble Machines. The announcement coincided with the company's debut at NVIDIA GTC 2026 in Booth #3303, with additional presence in the NVIDIA Booth #941. Wei Ding stated in the announcement that the company had shipped its first commercial robot deployments to a Fortune Global 500 industrial customer within 18 months of founding.[1][6]
In the rebrand, the company kept Moby as its core platform but introduced commercial offerings under the Noble Machines brand. The old undercontrol.ai domain was redirected via a 308 permanent redirect to noblemachines.ai.[6]
UnderControl AI was founded by four engineers with backgrounds spanning consumer electronics, aerospace, and academic robotics. As of 2026, the founding team continues to lead the company in their original roles.
| Name | Role | Prior background |
|---|---|---|
| Wei Ding | Co-founder and Chief Executive Officer | Apple (autonomous systems, around 7 years); textile-industry robotics in Shenzhen |
| Wenlong Ma | Co-founder and Chief Technology Officer | Ph.D. at Caltech; built one of the first running humanoid robots in the world during his doctoral work |
| Chris McQuin | Co-founder and Head of Hardware | Chief engineer of NASA's Valkyrie humanoid robot program at the Johnson Space Center |
| Wenda Wang | Co-founder and Head of AI | Led perception systems at Apple before founding UCR |
In an interview with Arm Newsroom, Ding described Arm's edge processors as roughly three to four times faster than competing architectures on the matrix vectorization workloads that dominate real-time humanoid control loops.[2]
Moby is Noble Machines' flagship bipedal humanoid robot. The platform is purpose-built for industrial work environments and emphasizes payload and durability over biomimicry, a design philosophy the company has publicly contrasted with competitors that prioritize visual human-likeness or top walking speed.[1][7]
The robot was built in roughly four months by the original four-engineer team and reached commercial deployment in approximately eight months. Moby uses a single NVIDIA Jetson Orin edge AI computer as its main compute platform, with STMicroelectronics STM32 microcontrollers (also Arm-powered) handling motor control in each actuator module.[2]
| Specification | Details |
|---|---|
| Form factor | Bipedal humanoid |
| Height | Approximately 170 cm |
| Weight | 72 kg (per UCR substack); 45 kg (per Humanoid.guide prototype listing) |
| Payload (max) | 60 lbs (27 kg) |
| Battery life | 5 hours per charge (around 5 to 6 hour shifts) |
| Average power consumption | About 200 W |
| Walking speed | Approximately 0.8 m/s (around 3 km/h) |
| Degrees of freedom | 34 total (10 in the hands) |
| Compute | Single NVIDIA Jetson Orin edge AI computer |
| Motor controllers | Arm-based STM32 microcontrollers |
| Operating system | Linux (Ubuntu) with ROS |
| Sensors | Lightweight IMU sensors (smartphone-grade orientation detectors) |
| Connectivity | Bluetooth, Wi-Fi |
| Frame materials | Composite polymer and aluminum |
| Ingress protection | IP20 |
| Development time | Around 4 months for the prototype; about 8 months to first deployment |
| Original team size | 4 engineers |
| Target industries | Manufacturing, logistics, construction, energy, mining, semiconductors |
Moby's quoted weight and walking speed differ slightly between the company's own substack post (72 kg) and the Humanoid.guide prototype directory (45 kg), reflecting different revisions of the platform during the stealth and post-stealth phases.[1][7]
Moby is designed for human workflows, meaning it can operate in environments and task sequences originally designed for human workers without requiring significant facility modifications. The robot can navigate steep inclines, stairs, scaffolding, narrow passageways, and debris-strewn outdoor terrain.[1][8]
Noble Machines offers Moby through two distinct commercial channels on its current website:
| Offering | Description |
|---|---|
| The Platform | Software toolkits and hardware access aimed at internal robotics teams at customer organizations who want to develop their own skills and integrations on top of Moby |
| The Service | Limited Robotics as a Service (RaaS) pilots operated directly by the Noble Machines field team, intended for customers who want a turnkey deployment |
UnderControl AI's technology stack combines NVIDIA's robotics simulation tools, Arm-based edge compute, and a proprietary AI pipeline focused on rapid sim-to-real transfer.
The whole-body control system coordinates Moby's legs, torso, arms, and hands to maintain balance while performing manipulation tasks, even on uneven surfaces, inclines, or while carrying payloads of varying weights. This contrasts with approaches that treat locomotion and manipulation as separate control problems.[1][8]
In the Arm Newsroom feature, Ding noted that Moby achieves its stability primarily through software and AI rather than expensive proprioceptive hardware, using inexpensive smartphone-grade IMU sensors and reinforcement-learning policies trained in simulation.[2]
Noble Machines' autonomy system is designed so that robots can learn new real-world skills in hours rather than months. The pipeline is built on top of NVIDIA Isaac Lab and Isaac Sim for simulation, with Isaac GR00T used to generate synthetic motion and manipulation data for training new skills. NVIDIA X-Mobility handles end-to-end navigation with world modeling.[1][6]
The company describes its training pipeline as a four-stage process:
| Stage | Description |
|---|---|
| Real2Sim | Embedding measured actuator and contact dynamics from physical hardware into the simulator |
| Sim2Sim | Cross-platform evaluation between different simulators to test robustness of learned policies |
| Sim2Real | Domain randomization combined with adaptive learning to bridge the reality gap |
| Real2Real | Tolerance to hardware variation between individual robot units |
This pipeline is credited by the company with achieving a roughly 95% sim-to-real deployment success rate, meaning policies trained in simulation perform on the real robot at expected levels without extensive on-hardware tuning.[1]
Moby supports three training modalities for operators on the factory floor:
| Modality | Description |
|---|---|
| Language-based instructions | Operators can describe a new task in natural language and the robot interprets it through a large language model component |
| Physical demonstrations | A human operator can physically guide the robot's arms and body through a task one or more times, and the robot learns the policy |
| Gesture-based teaching | Operators can use hand gestures and pointing to indicate objects, locations, and sequences |
This multi-modal approach is intended to reduce the technical expertise required to program the robot for new tasks, so that factory workers and operators on the line can teach the robot directly without writing code.[1][8]
UnderControl AI develops its AI-driven robots on the NVIDIA Isaac platform and runs the entire onboard software stack on a single Jetson Orin module. The company uses NVIDIA CUDA for development and deployment workflows. Wei Ding has stated that one Jetson Orin module provides enough compute for Moby's full control stack including perception, planning, and whole-body control inference.[1][2]
Moby uses Arm processors at two layers of the system. The main compute is an Arm Cortex-A78 class CPU integrated on the NVIDIA Jetson Orin board, which runs the autonomy stack and AI inference. The motor controllers in each actuator module use Arm-powered STM32 microcontrollers from STMicroelectronics. According to Ding, Arm was selected because it offered the best combination of energy efficiency, cost-effectiveness, and performance for edge robotics, particularly for the matrix vectorization workloads at the core of real-time control.[2]
Noble Machines targets what it calls "4D jobs," the dull, dirty, dangerous, and declining categories of industrial work where employers struggle to recruit and retain human workers. The company has described its approach as building robots for "the world's most dangerous jobs" and "the hazardous, physically demanding industrial tasks that keep the world moving."[1][2][9]
| Industry | Example tasks |
|---|---|
| Manufacturing | Assembly, material handling, and quality inspection on the production line |
| Logistics | Warehouse operations, package handling, and intra-facility transport |
| Construction | On-site material transport, tool delivery, and component placement |
| Energy | Inspection and maintenance in hazardous environments, including offshore platforms |
| Mining | Underground inspection, material handling, and equipment tending |
| Semiconductors | Clean-room operations and equipment tending |
As of the March 2026 stealth exit, Noble Machines has publicly named several technology and channel partners. The company has not disclosed the name of its first Fortune Global 500 customer.[1][6]
| Partner | Role |
|---|---|
| NVIDIA | AI compute (Jetson Orin), simulation stack (Isaac Sim, Isaac Lab, Isaac GR00T), navigation (X-Mobility), and CUDA |
| Arm | Processor architecture for main compute and motor controllers |
| ADLINK | Edge computing platforms |
| Schaeffler | Motion technology and humanoid robotics components |
| Solomon | Factory floor application integrations |
| Diverge | Listed as a partner on the noblemachines.ai site |
| STMicroelectronics | STM32 motor-controller microcontrollers used in each actuator module |
A Fortune Global 500 customer quoted by the company in March 2026 said that Noble Machines was the only company shipping humanoid robots with the payload capacity the customer needed for its industrial use case. ADLINK's general manager praised the company's whole-body control and autonomy stack in the same announcement.[6]
Noble Machines has raised an undisclosed seed round prior to its stealth exit. Investors named in Crunchbase and Tracxn listings include South Park Commons (an early-stage accelerator and community in San Francisco), UP Partners, Centre Street Partners, Nirman Ventures, Oakseed Ventures, and Paperjet Ventures. Specific dollar amounts and dates have not been publicly disclosed.[10]
In interviews and on its own blog, UnderControl AI has articulated a design philosophy that contrasts with humanoid robot makers that emphasize cinematic walking speed or visual likeness to a human. The company calls its approach "anti-human" in the sense that it does not optimize Moby's gait or appearance to match a human; instead, it optimizes for safety, energy efficiency, payload capacity, and useful agility on industrial terrain.[1][7]
Human-like motions occasionally emerge as a natural consequence of optimizing for energy efficiency and stability on bipedal hardware, but the company has been clear that biomimicry is not the design goal. The phrase "application-essential performance" recurs in the company's public materials, referring to the principle that any feature that does not directly help the robot complete its industrial task should be removed.[1][7]
The company received early industry attention before its public stealth exit. The Humanoid Hub described visiting the UCR lab in Santa Clara as "the kind of startup story you love to hear," highlighting the small team and rapid prototype timeline. The Robot Report and Humanoids Daily both covered the March 2026 stealth exit, and the Arm Newsroom published a profile of the company's Arm-powered humanoid robots in 2025.[1][2][5]
At NVIDIA GTC 2026, Noble Machines demonstrated Moby in both its own booth and within the main NVIDIA booth, and it was featured in the AI Insider's coverage of new humanoid platforms launching that year.[6]