NVIDIA Corp. (NVDA) is focusing on enhancing the operational safety of humanoid robots in human environments. The world's highest-valued chipmaker asserts that for robots to be genuinely trusted for close collaboration with people, they must possess the capability for instantaneous, millisecond-level decision-making.
The company announced on Monday that it will provide software and semiconductor solutions to enable humanoid robots to enter workplaces and engage in real interactions with humans—including physical contact when necessary. Its Halos software, derived from autonomous driving systems, will act as the robot's "perception hub," significantly boosting its situational awareness of the surrounding environment.
NVIDIA and its Silicon Valley peers are racing to position robotics as the next major frontier in artificial intelligence. Technology executives predict that humanoid robots could evolve into a vast market encompassing billions of units. According to estimates from Barclays, humanoid robot technology—featuring bionic designs that mimic the human head, torso, and limbs—could generate annual revenue of $200 billion by 2035.
However, the industry still needs to overcome a series of safety challenges. Current safety systems, which force robots to stop or slow down when they might come into contact with a human, ensure safety but severely hamper productivity. They also hinder interactive operations such as delivering items or collaboratively moving heavy objects.
"Traditional robot safety logic essentially involves putting them in a cage or using sensors to detect obstacles and then making the robot stop abruptly. But that's far from sufficient for humanoid robots," noted Amit Goel, Senior Director of Product Management at NVIDIA.
NVIDIA will provide technical support for humanoid robots like Agility Robotics' Digit, enabling them to make autonomous decisions based on real-time analysis of their surroundings. NVIDIA stated that this requires hardware capable of supporting safety perception functions, and the company will offer Halos as an operating system running on its IGX Thor hardware platform. Additionally, robots will be connected to external sensors.
For instance, an autonomous forklift could access cameras within a warehouse to perceive conditions around blind corners in advance, allowing it to decide whether to maintain full speed or slow down to avoid obstacles.
As the world's most valuable company, NVIDIA currently derives most of its revenue from AI data center chips but is gradually expanding into other areas—offering some technologies even for free—to broaden the applications of artificial intelligence.
To further accelerate the deployment of robots, NVIDIA has established a dedicated laboratory where robot manufacturers and clients can conduct safety testing before applying for certification from regulatory bodies. NVIDIA engineers will handle pre-inspection work and assist with engineering optimizations as needed.
The safety constraints for humanoid robots are fundamentally different from those for autonomous vehicles. The latter only need to avoid collisions with people or objects, whereas the former require a much higher degree of flexibility.
"The safety design must be far more advanced because you need to decide, based on environmental perception, what objects can be touched, moved, or have force applied to them, and to what degree, all in real-time. These forces cannot be too small—you can't solve the problem by making the robot too weak because a robot that fragile simply can't perform useful work," explained Pras Velagapudi, Chief Technology Officer at Agility Robotics.
Although engineers continue to face the technical challenges of creating "artificial humans," the market is large enough for the industry to learn and iterate through phased development. Velagapudi pointed out that domestic care robots will be a later-stage product, which must be built upon the mature application of factory robots.
"We are starting with warehousing and logistics, which is one of the most structured environments, yet already has a multi-billion dollar addressable market. In the future, we can expand into retail, healthcare, and even construction. The difficulty increases with each field, but each field itself is a massive industry capable of supporting the deployment of a vast number of robots," he stated.
Comments