Humanoid Builds HMND 01 Alpha in 7 Months Using NVIDIA Robotics Stack
London-based startup Humanoid moved from concept to a functional alpha prototype of its HMND 01 robot in seven months, compressing a development cycle that typically takes up to two years.
London-based robotics startup Humanoid has compressed the traditional hardware development timeline by moving from concept to a functional alpha prototype of its HMND 01 system in just seven months.
The milestone stands in contrast to the typical 18 to 24 months required to develop comparable humanoid or industrial robotic platforms, highlighting how simulation-first development and edge AI are reshaping robotics engineering.
The HMND 01 Alpha program includes two robot variants: a wheeled platform designed for near-term industrial deployment and a bipedal system intended primarily for research and future service or household applications.
Both platforms are currently undergoing field tests and proof-of-concept demonstrations, including a recent industrial evaluation with automotive supplier Schaeffler.
At the center of Humanoid’s accelerated development cycle is a tightly integrated software and hardware stack built on NVIDIA robotics technologies.
Edge Compute and Foundation Models at the Core
The HMND 01 Alpha robots use NVIDIA Jetson Thor as their primary edge computing platform. By consolidating compute, sensing, and control onto a single high-performance system, Humanoid simplified internal architecture, wiring, manufacturability, and field serviceability.
Jetson Thor allows the robots to run large robotic foundation models directly on-device rather than relying on cloud processing. This enables real-time execution of vision-language-action models that support perception, reasoning, and task execution in dynamic environments.
Humanoid reported that training these models using NVIDIA’s AI infrastructure has reduced post-training processing times to just a few hours. This faster turnaround significantly shortens the loop between data collection, model refinement, and deployment on physical robots, allowing the company to iterate at software speed rather than hardware speed.
Simulation-First Development and Hardware Optimization
Humanoid’s workflow is built around a simulation-to-reality pipeline using NVIDIA Isaac Lab and Isaac Sim. Engineers use Isaac Lab to train reinforcement learning policies for locomotion and manipulation, while Isaac Sim provides a high-fidelity environment for testing navigation, perception, and full-body control.
Through a custom hardware-in-the-loop validation system, Humanoid created digital twins that mirror the software interfaces of the physical robots. This allows middleware, control logic, teleoperation, and SLAM systems to be tested virtually before deployment on real hardware. According to the company, new control policies can be trained from scratch and deployed onto physical robots within roughly 24 hours.
Simulation also plays a direct role in mechanical engineering decisions. During development of the bipedal robot, Humanoid evaluated six different leg configurations in simulation, analyzing torque requirements, joint stability, and mass distribution before committing to physical prototypes.
Engineers also optimized actuator selection, sensor placement, and camera positioning using simulated perception data, reducing the risk of blind spots and interference in industrial settings.
These physics-based simulations contributed to the robots’ performance during early industrial trials and helped avoid costly redesigns later in the development cycle.
Toward Software-Defined Robotics Standards
Humanoid views HMND 01 as part of a broader shift toward software-defined robotics. The company is working with NVIDIA to move away from legacy industrial communication standards and toward modern networking architectures designed for AI-enabled robots.
“NVIDIA’s open robotics development platform helps the industry move past legacy industrial communication standards and make the most of modern networking capabilities,” said Jarad Cannon, chief technology officer of Humanoid.
He added that the company is collaborating on a new robotics networking system built on Jetson Thor and the Holoscan Sensor Bridge, with the goal of enabling more flexible and scalable robot architectures.
Founded in 2024 by Artem Sokolov, Humanoid has grown to more than 200 engineers and researchers across offices in London, Boston, and Vancouver. The company reports 20,500 pre-orders, six completed proof-of-concept projects, and three active pilot programs.
While the bipedal HMND 01 remains focused on research and long-term service robotics, the wheeled variant is positioned for near-term industrial use. Humanoid’s strategy emphasizes early deployment in operational environments to gather real-world data and continuously refine its software-defined architecture, signaling a shift in how humanoid and industrial robots are developed and brought to market.
