Hillbot Teams Up With NVIDIA Cosmos Video Generation To Improve Robotics Training

Hillbot trains robots using AI-generated videos from NVIDIA, speeding up how quickly robots learn new tasks while cutting training costs.

The robotics company uses NVIDIA's Cosmos platform to create realistic video clips of different environments. This helps robots learn tasks like moving drinks from carts to fridges without needing physical practice spaces.

Training robots usually requires real-world practice, which takes time and money. Companies often build multiple test areas to teach robots different tasks.

Hillbot tackles this problem by using AI to make videos of various kitchen layouts and fridge designs. These videos help create virtual training worlds through the company's Sapien and ManiSkill tools.

The process starts with NVIDIA NeMo Curator, which prepares training videos. Cosmos then generates clips showing different kitchen setups that robots might encounter.

These virtual spaces let robots practice millions of times in different situations. This method costs less than building physical test areas and helps robots learn faster.

The company says robots trained this way perform well in real kitchens, adapting to different layouts and fridge types they encounter.

This approach marks a shift in robot training methods, showing how AI-generated content can make robots smarter and more adaptable while saving time and money.