Humanoid follows a simulation-first approach using NVIDIA Isaac Lab and Isaac Sim. | Credit: Humanoid
By moving from concept to a functional alpha prototype of its HMND 01 system in seven months, London-based startup Humanoid is attempting to compress the traditional robotics hardware development cycle of 18 to 24 months.
The company’s HMND 01 Alpha robots, which include both wheeled industrial and bipedal research platforms, are currently undergoing field tests and proof-of-concept demonstrations.
Central to this development speed is an integrated software and hardware stack provided by NVIDIA.
Edge compute and foundation models cut complexity
The HMND 01 Alpha uses NVIDIA Jetson Thor as its primary edge-computing platform. For developers, the shift to Thor represents a move toward consolidating the robot’s internal architecture.
By running large-scale robotic foundation models directly at the edge, Humanoid claimed that it has reduced the complexity of its system wiring and simplified field serviceability.
The compute capacity of the platform allows for the execution of vision-language-action (VLA) models on the device. Humanoid reported that using NVIDIA’s AI infrastructure for training these models has reduced post-training processing times to several hours, accelerating the iteration loop between data collection and deployment.
Humanoid has a simulation-first pipeline
Humanoid’s development workflow relies on a simulation-to-reality (Sim2real) pipeline built on NVIDIA Isaac Lab and Isaac Sim. Its engineering team uses Isaac Lab to train reinforcement learning (RL) policies for locomotion and manipulation.
This virtual training environment allows the team to develop and deploy a new policy from scratch onto physical hardware in approximately 24 hours.
To bridge the gap between simulation and the physical world, the company developed a custom hardware-in-the-loop (HIL) validation system.
By creating digital twins that utilize the same software interfaces as the physical HMND 01 robots, engineers can test middleware, control systems, and teleoperation setups in a virtual environment before running them on hardware.
This environment is also used to validate simultaneous localization and mapping (SLAM) and navigation policies.
Engineers use physics-based hardware optimization
Simulation is used as a tool for mechanical engineering rather than just software validation. During the design of the bipedal robot, Humanoid’s engineers evaluated six different leg configurations within Isaac Sim.
By analyzing torque requirements, mass distribution, and joint stability in the virtual environment, the company said its team optimized actuator selection and joint strength before manufacturing physical prototypes.
This approach enabled the optimization of sensor and camera placement based on simulated perception data, reducing the risk of blind spots or interference in real-world industrial environments.
The ability to analyze forces and motion virtually contributed to the performance of the robots during a recent proof of concept with automotive supplier Schaeffler.
Goal is to transition to software-defined standards
Humanoid said one of its core concepts is to transition away from legacy industrial communication standards toward modern networking. The company is collaborating with NVIDIA to develop a robotics networking system.
“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.
“We’re currently working closely with NVIDIA and other partners on a new robotics networking system built on Jetson Thor and the Holoscan Sensor Bridge,” he added. “We believe this co-developed open network standard for AI-enabled robots could make a big impact across the industry. Together, we can open the way for software-defined robots.”
Organization scales with HMND 01 deployment
Founded in 2024 by Artem Sokolov, Humanoid now employs over 200 engineers and researchers across offices in London, Boston, and Vancouver. While the bipedal robot remains a research and development tool for future household applications, the wheeled HMND 01 variant is intended for immediate industrial use.
The company currently reports 20,500 pre-orders and has three active pilot programs. Humanoid’s focus remains on bringing these systems into operational environments early to gather performance data and iterate on the software-defined architecture.
Comparing HMND 01 Alpha wheeled and bipedal versions
| Specification | HMND 01 Alpha (Wheeled) | HMND 01 Alpha (Bipedal) |
|---|---|---|
| Primary Deployment | Industrial Logistics & Warehousing | Service R&D & Household Apps |
| Locomotion | 4-Wheel High-Stability Base | 2-Leg Dynamic Balance |
| Degrees of Freedom | 29 DoF (Torso + Arms) | 29 DoF (Full Body) |
| Compute Engine | NVIDIA Jetson Thor | NVIDIA Jetson Thor / Orin AGX |
| Max Speed | 7.2 km/h (4.4 mph) | 5.4 km/h (3.4 mph) |
| Payload Capacity | 15 kg (Bimanual) | 15 kg (Bimanual) |
| Vision System | 360° RGB + Dual Depth Sensors | 6x RGB + Dual Depth Sensors |
| Power Management | 8 Hours (Auto-Charging) | 3–4 Hours (Swappable Battery) |
| Height | 220 cm | 179 cm |
| Dev Cycle (Alpha) | 7 Months | 5 Months |
Tell Us What You Think!