Humanoid Expands Nvidia Partnership for Onboard Robot AI

LONDON -- Robotics company Humanoid is expanding its collaboration with Nvidia to develop an onboard computing platform for its next generation of humanoid robots, combining AI processing, sensor data and robot control in a single architecture.

The London-based company announced the expanded collaboration Oct. 21, saying it is integrating Nvidia's Jetson Thor computing platform, Isaac OS, Holoscan Sensor Bridge and Holoscan software development kit.

The technology is intended to let Humanoid's robots process camera and other sensor data, run AI models and coordinate movements without relying on external computing systems for those workloads.

Humanoid said its transition to Jetson Thor has allowed it to consolidate two onboard computers into one while maintaining computing capacity. The platform's shared CPU and GPU memory also reduces data transfers between processors, potentially lowering latency.

The company is integrating Holoscan Sensor Bridge to move camera data directly into memory accessible by the graphics processor, reducing unnecessary copying and CPU workload.

Isaac OS provides a runtime environment for coordinating multiple robotics workloads, while the architecture supports network bandwidth of up to 100 gigabits per second for sensor information and control signals, according to Humanoid.

The companies have already worked together on robot training and simulation. Humanoid uses Nvidia Isaac Lab and Isaac Sim to train and test AI models in simulated environments before transferring them to physical machines.

The expanded collaboration focuses on the computing architecture inside the robots rather than introducing a new humanoid model.

Humanoid has not disclosed a release date for its next-generation robots or published independent benchmarks showing improvements in processing speed, power consumption or real-world task performance.