Humanoid Robotics and NVIDIA Partner to Accelerate Humanoid Robot Development

London-based robotics company Humanoid has announced a collaboration with NVIDIA to advance the development of its humanoid robot. The partnership centres on integrating NVIDIA’s accelerated computing, simulation tools and edge processing technologies to address key challenges in robotics, including rapid prototyping and the creation of robust artificial intelligence. Humanoid, founded in 2024 and now operating from offices in London, Boston and Vancouver, aims to reduce development cycles to six weeks by utilising NVIDIA’s platforms for virtual testing and refinement, ultimately targeting applications within manufacturing, logistics and retail. This collaboration seeks to accelerate the deployment of commercially viable and reliable humanoid robots capable of operating effectively in complex real-world environments.

Accelerated Development Through Simulation

Humanoid is substantially accelerating its robotic development cycles through a ‘sim-first’ approach, leveraging NVIDIA’s Isaac Sim and Omniverse platforms. This methodology enables rapid iteration of robot designs within virtual environments, aiming to reduce prototyping timelines to six weeks. By conducting extensive validation of numerous design permutations in simulation, Humanoid optimises designs and mitigates potential issues prior to physical manufacturing, streamlining the development process. This strategy is further enhanced by the application of reinforcement learning (RL) and data generation techniques to train robust robot control policies, encompassing both body control and advanced manipulation capabilities, preparing the robots for complex tasks in real-world environments.

Humanoid further advances its robots’ capabilities through the training of sophisticated Vision-Language-Action (VLA) models. These models integrate visual perception, natural language understanding, and action planning, enhancing the robot’s agility and improving its capacity for intuitive interaction with humans. By integrating VLA models, Humanoid aims to create robots capable of reliably performing tasks across a diverse range of environments and applications.

Scaling Robotics with Massive Data

Central to Humanoid’s strategy is the application of large-scale data generation and simulation, facilitated by NVIDIA’s Isaac Sim and Omniverse platforms. This ‘sim-first’ development process significantly reduces prototyping cycles to an estimated six weeks, allowing for rapid iteration and refinement of robot designs through virtual validation before physical manufacturing. The scale of these simulations is substantial, providing the necessary data to ensure the robots can reliably perform their designated functions across a range of conditions and scenarios.

Furthermore, Humanoid leverages NVIDIA’s computational resources to train sophisticated Vision-Language-Action (VLA) models, integrating visual perception, natural language understanding, and action planning. This advanced AI training expands the robot’s autonomous capabilities, enabling it to operate reliably and adapt to diverse, unpredictable situations. The integration of the upcoming NVIDIA Thor platform provides Humanoid’s robots with high-performance edge computing capabilities, supporting real-time perception, complex decision-making, and dynamic interactions, all essential for effective operation in varied environments.

Advancing AI for Autonomous Operation

Humanoid is actively developing autonomous capabilities in its humanoid robots through a strategy centred on accelerated artificial intelligence and advanced simulation. The company employs NVIDIA’s Isaac Sim and Omniverse platforms to facilitate a ‘sim-first’ development process, significantly reducing prototyping cycles to an estimated six weeks, allowing for rapid iteration and refinement of robot designs through virtual validation before physical manufacturing.

A core component of this advancement lies in the application of massive-scale simulations utilising reinforcement learning (RL) to develop robust body controllers and sophisticated manipulation policies. This process prepares the robots for complex, real-world tasks by enabling them to learn and adapt through simulated experience, improving their performance and reliability in dynamic environments. Furthermore, Humanoid is training Vision-Language-Action (VLA) models to enhance robot agility and improve human-robot interaction, allowing the robots to interpret instructions, recognise objects, and execute tasks with greater precision and intuitiveness.

The integration of the forthcoming NVIDIA Thor platform provides Humanoid’s robots with high-performance edge computing capabilities, supporting real-time perception, complex decision-making, and dynamic interactions. By processing data locally, the robots can respond quickly to changing conditions and operate independently, reducing reliance on external connectivity and improving overall system responsiveness.