Physical AI in manufacturing: Seeing beyond the demonstration

Author:
Katie Windley
Marketing Manager

Robots can run, navigate unfamiliar environments, and respond to increasingly complex instructions. Those demonstrations can make it tempting to focus on what the latest technology is capable of doing.

For manufacturers, there is another set of questions to answer. What problem does the robot solve? Why does the application require emerging robotics instead of conventional automation? How consistently can it perform the task? And what happens when something goes wrong?

In a recent episode of People B4 Machines, host Amanda Cupido spoke with robotics evangelist, angel investor, and content creator Lukas M. Ziegler about physical AI, world models, and how manufacturers can evaluate increasingly capable robotics without losing sight of the application.

Key Takeaways

Listen to the full conversation

Hear Lukas M. Ziegler discuss physical AI, world models, the changing role of robotics integrators, and the human questions surrounding increasingly capable machines at peopleb4machines.com.

Give robots an understanding of the physical world

Generative AI largely operates in a digital environment. A robot has a different challenge: it needs to understand and act on the physical world around it.   

Ziegler uses a chair as a simple example. A robot needs to recognize that the object is a chair, but it also needs spatial information about the object and the actions that are possible.

• Can it be moved?
• Where is it located?
• How can the robot physically interact with it?

World models are one approach to giving robots this type of reasoning and spatial understanding. Ziegler expects these capabilities to make robots increasingly autonomous and able to perform tasks without every action being explicitly programmed.  

That doesn’t mean manufacturing will suddenly have its own equivalent of the ChatGPT moment. Physical environments introduce variables that digital AI doesn’t have to contend with, and the information required to train robots is considerably different.  

_{Physical AI could help robots understand objects, spaces, and how to interact with them.}

Rethink how robots are integrated

If robots can reason through more of a task, the work required to deploy them could change too.  

Ziegler uses the term “neo-integrator” to describe what he sees as a potential next iteration of the traditional robotics integrator. Rather than writing explicit instructions for every action, an integrator could increasingly define the task and operating framework, with tools such as world models and teleoperation handling other aspects of execution.

“You gotta be able to unlearn very quickly these days,” Ziegler says. His prediction is deliberately forward-looking: he expects software to remain central to robotics, while the way integrators interact with and create that software changes.  

He also sees limits. World models can be probabilistic, while manufacturing environments often require predictable behaviour for safety, risk analysis, and certification. Ziegler expects deterministic logic and established control systems to continue playing a role alongside newer approaches. 

Put the application before the technology

For manufacturers trying to understand which emerging robotics technologies deserve attention, Ziegler’s experience as an angel investor provides a useful framework.

“Can you name the customer who is using your robots? Can you tell me what task your robot is doing and why this task is being done by a robot 2.0… rather than a simple machine?” he asks.

From there, Ziegler looks at cycle time, performance, technology readiness, edge cases, and how well a robot recovers when something goes wrong. Those questions shift the conversation away from what a robot can demonstrate toward what it can reliably accomplish.

An emerging technology becomes more compelling when it enables manufacturers to address work that conventional automation couldn’t. Starting with the application also helps determine whether the added complexity of a newer robotics platform is justified.

_{The value of emerging robotics depends on the task, performance, and reliability.} 

Prepare people to work closer to robots

Technical capability is only one part of introducing a new robot into a manufacturing environment. Employees also need to understand what the technology is doing and how they’ll interact with it.

Ziegler expects this to become particularly important with humanoid robots. “I think the best answer to convince people for automation in robotics is actually using it in the most kind of necessary moments,” he says. In his view, adoption will be gradual, with manufacturers needing to demonstrate where robots add value and help employees.

Current limitations are part of that discussion. Ziegler points specifically to the unpredictability of bipedal locomotion and the risks associated with a large machine losing its balance around people. For now, he sees significant work ahead before this type of human-robot interaction becomes routine.

Explain what the technology can see

As robots gain greater awareness of their surroundings, cameras and sensing technologies introduce another employee concern: surveillance.

Technical safeguards can help address privacy, but Ziegler argues that manufacturers also need to communicate what the technology is doing. Employees should understand why a camera is present, how its information is being used, what the technology can and cannot do, and what privacy safeguards are in place.

That communication becomes increasingly important as automation systems collect more information about the environments in which they operate. The people working around those systems may not share the same familiarity with, or enthusiasm for, the technology as the teams deploying it.

_{As robots become more aware of their surroundings, employees need to understand what the technology can see and how information is used.} 

Explore the possibilities

Physical AI could open new opportunities for automation in manufacturing. Learn how Eclipse Automation can help you evaluate emerging robotics, identify where they fit, and bring them into the production environment.