Nvidia's Big Bet on Physical AI Aims for Safer Robotaxis, Humanoid Robots
Full-stack safety solution for physical AI is being used by robotics companies.
TL;DR
- Nvidia is developing a full-stack safety system called Halos for physical AI technologies like robotics and self-driving cars.
- The Halos system, launched in 2025 for autonomous vehicles, was expanded to robotics in June 2026.
- It aims to reduce the risk of machines harming people by providing hardware and software tools for safety guardrails.
- The physical AI business is already a significant revenue driver for Nvidia.
- The safety system includes hardware like the IGX Thor computing module and software for constant monitoring and isolation of critical workloads.
- It also incorporates the Holoscan Sensor Bridge for detecting corrupted sensor data and simulation tools for virtual testing.
- Adapting safety for robotics required accommodating diverse definitions of safety compared to the more standardized automotive sector.
- Companies like Agility Robotics and Boston Dynamics are incorporating Nvidia Halos into their robots.
- The push for safer robots is essential for enabling them to perform multiple tasks in various environments.
- Nvidia's Halos aims to allow robots to operate more freely without physical barriers or isolated workstations.