People have been dreaming about intelligent humanoid robots since the introduction of the beloved family maid Rosey on The Jetsons in the early 1960s. However, bringing robots into everyday life has been slow and challenging. Like any new technology application, it takes an entire ecosystem to evolve to make the application practical, especially for consumer use. In the past 60 years, we have seen the evolutions in semiconductors, sensors, wireless interfaces, hydraulics, electrics motors, positioning solutions, gyroscopes, accelerometers, and battery technology, which are all required in a complex robotic system. During that time, robots have advanced significantly to do complex tasks in manufacturing, warehouses, operating rooms, and even our homes. Yes, the robotic vacuum is still a robot. But they have never been truly intelligent. That last piece in the puzzle has emerged with agentic AI – the ability to reason and initiate action autonomously based on that reasoning.
Disclosure: My company, Tirias Research, has consulted for AMD, Intel, Nvidia, NXP, Qualcomm and other companies mentioned in this article.
Generative AI enabled the ability to think and develop the best response based on the available training data, but agentic AI enables the ability to consider all options based not just on the training data, but all available data, such as sensor data about the current environment. This may be the final piece of the puzzle that launches the era of intelligent robots, including humanoid robots.
At Computex in Taiwan last year and CES earlier this year, Nvidia CEO Jenson Huang replicated the iconic Ironman scene where humanoid robots rise onto the stage. He referred to this as a new transition in AI to “physical AI,” and it would be hard to argue with that assessment. There are now advanced processing solutions available from dozens of vendors, such as AMD, Intel, Nvidia, and Qualcomm; advanced sensors from the likes of ams OSRAM, Analog Devices, Honeywell, NXP, ON Semiconductor, and STMicroelectronics to name a few; wireless communications and positioning solutions from companies like MediaTek, NXP, Qualcomm, Silicon Labs, and TI; not to mention the hundreds of other system components and component vendors that make up a rich but essential ecosystem to enable the physical AI era.
In addition, we have advanced robotic training, simulation, and management platforms. The most notable being the platforms from Nvidia. Nvidia has spent more than a decade developing everything from hardware with its Jetson edge platforms to its cloud-based GPU training solutions, and software solutions, such as Isaac sim, Omniverse, Cosmos, GR00T, and Fleet Management for the simulation, training, and management of a single robot to a fleet of robots. We will probably hear more about robotics and physical AI at the 2025 GPU Technology Conference (GTC) in San Jose this week.
Over the past year, I have seen dozens of humanoid robots at the major technology tradeshows, and I have never been more impressed. It was one thing to watch a robot from Boston Dynamics transverse an obstacle course, but it is another when a robot walks up to you, shakes your hand, and starts talking to you. COVID and continued labor shortages have driven a push for AI, autonomous systems, and robots across industry segments, but with agentic AI we are entering the robotic, or what Jenson Huang calls the physical AI, era. It may be difficult to call 2025 the year of the robot because we have so much more innovation coming, but between now and 2030, intelligent robotics will start to reach almost every aspect of society.
