NVIDIA Introduces HOVER, a 1.5 M Parameter Neural Network for Humanoid Robotics

Researchers at NVIDIA have developed HOVER (humanoid versatile controller), a 1.5 million parameter neural network designed to coordinate the motors of humanoid robots for locomotion and manipulation.

“Not every foundation model needs to be gigantic. We trained a 1.5M-parameter neural network to control the body of a humanoid robot,” said Jim Fan, senior research manager and lead of embodied AI (GEAR Lab) at NVIDIA.

Fan added that this model captures the subconscious processes involved in human movement, allowing robots to execute complex tasks without extensive programming. “It takes a lot of subconscious processing for us humans to walk, maintain balance, and maneuver our arms and legs into desired positions,” he explained.

HOVER was trained in NVIDIA Isaac, a GPU-powered simulation suite that accelerates physics simulations by 10,000 times faster than real time. Fan said the model underwent a year of virtual training in a simulated environment, taking only about 50 minutes of real-world time on a single GPU. He explained that this efficiency allows the neural network to transfer seamlessly to real-world applications without requiring fine-tuning.

HOVER can respond to various high-level motion instructions, including commands for head and hand poses using XR devices like the Apple Vision Pro, whole-body poses from motion capture or RGB cameras, joint angles from exoskeletons, and root velocity commands from joysticks, according to Fan

Moreover, the model provides a unified interface for controlling robots with different input devices, facilitating the collection of teleoperation data for training purposes.

It also integrates with an upstream Vision-Language-Action model to convert motion instructions into low-level motor signals at high frequency. Compatible with any humanoid that can be simulated in Isaac, HOVER enables users to easily bring their robots to life.

Earlier this year, NVIDIA also announced Project GR00T, a general-purpose foundation model for humanoid robots.

Robots powered by GR00T, short for Generalist Robot 00 Technology, are engineered to understand natural language and mimic human movements by observing actions. This allows them to quickly learn coordination, dexterity, and other skills required to navigate, adapt, and interact effectively in the real world.

The post NVIDIA Introduces HOVER, a 1.5 M Parameter Neural Network for Humanoid Robotics appeared first on AIM.

Follow us on Twitter, Facebook
0 0 votes
Article Rating
Subscribe
Notify of
guest
0 comments
Oldest
New Most Voted
Inline Feedbacks
View all comments

Latest stories

You might also like...