Let's get physical! How Hitachi Vantara is getting ready for the next manifestation of AI

At the moment agentic AI is at the height of its hype cycle and, consequently, the market is focused on the application of AI in the virtual world to enable intelligent agent-driven planning, decision making, workflow automation, and process optimisation. This has meant that currently the AI spotlight is on enterprise back and front office business systems – the suited side of the organization, rather than on the factory floor. 

However, there are rumblings that the market is about to get a more physical manifestation of cutting-edge AI. Of course, automation has been used in industrial settings for decades now, and, indeed, machine “vision” applications, providing the ability to process images and video feeds, requires AI technology. 

But is “physical AI” about to grab its share of the market limelight? I recently spoke with Jason Hardy, CTO of AI at Hitachi Vantara to find out if there is more to this next AI phase than autonomous cars.

What is physical AI?

Physical AI is where AI meets IoT, where machines will be able to act in the physical world in real-time. Physical AI has the physicality of robotics combined with a better understanding of the world around the robot, so that it can engage with the physical space. Hardy explains:

Physical AI is a move beyond the helper robotic, to the creation of a collaborative bridge between humans and robots. It will transform manufacturing to a point where robots can manage things that humans are doing and become an extension of the workforce. It takes a lot of skills that we are building in agentic AI and takes that agency into the physical world to improve the efficiency of the human workforce. Transporting things around (aka autonomous vehicles) is at the lower end of this capability.

We are looking at how we can create robotics that are skilled at an individual task. In a manufacturing line this could involve interacting with different materials or it could be the ability to introduce robotics into more manual tasks where they can move continuously as a human operates something.

We have already been using perception AI training so that, say, robots can identify damage in a can and discard it. With physical AI, robots can pick up a product and weld it in place and create something. This requires a lot more dynamic understanding of the complex physical environment as the robotic operates tools, understands materials and manoeuvres in the physical space.

He adds:

Digital twins are very important for this because they enable robots to adapt to the world around them. For example, you could create a customer service robot that you could ask a question such as, "Where do I go to get lunch?', and then it could take you to the place you need to be at, thus providing the answer in physical movement.

How is Hitachi Vantara preparing?

Hitachi Ltd, Hitachi Vantara’s parent company has a heavy industrial presence in sectors that will benefit from the physical AI world. In particular, Hitachi Vantara works closely with Hitachi Rail and Hitachi Energy and is currently learning the capabilities and understanding the dynamics of physical AI. 

The digital twin expertise Hitachi is building on NVIDIA’s Omniverse platform is crucial to this development of physical AI as it provides the ability to power training and operating spaces, because as Hardy comments:

You need enormous compute power for this to generate the photo-realistic rendering that needs to be performed in real-time in order train and build physical AI behaviors.

 Hitachi Vantara is working closely with its JR Automation robotics subsidiary to evolve the company’s AI analytics platform, Hitachi IQ, to train and design these environments.

Hardy continues:

As this technology evolves customers can plan for a better warehouse, assembly line and so on. This planning can be done via simulation, creating more efficiencies beyond those provided by automation and robotics. The benefits will be seen first in the manufacturing side, in railways and power grids via simulation and Hitachi will provide the telemetry and data to drive better outcomes.

We are seeing the benefits already in how we build our own products in Hitachi Rail– using digital twins we build to better plan and create awareness of sensors, which then enables more real-time awareness of conditions within the train and on the rail network via the carriage physical AI. This will vastly improve the reliability and predictability of health and safety on rail.

In manufacturing physical AI will enable us to build better manufacturing processes by taking robotics to the next level, training robots about multiple conditions in an environment as well as providing the ability for them to learn while operating.

My take

The idea of a customer service robot that is much more intelligent than, say, the mobile information kiosks in large spaces that exist today, is attractive. However, we return to the perennial AI question, how quickly can ROI on infrastructure spend and development of new products for physical AI be evidenced? The investment required is significant and returns will take time, which is why physical AI may not become a mainstream market technology very quickly. However, Hitachi is in a stronger position than many players to demonstrate the benefits of this approach to clients because of its application in its own industrial subsidiaries.