AI in robotics can plug persistent manufacturing gaps

Artificial intelligence (AI) has brought significant advancements in robotics that have potential to plug persistent gaps in manufacturing, specifically, the lack of manpower resources and customised solutions.

No longer rigid, preprogrammed automation, robotics are now software-defined systems capable of operating beyond controlled environments, said Charlie Dai, vice president and principal analyst at Forrester.

“Physical AI enables greater versatility, multifunctionality, and resilience, while fleets of robots increasingly coordinate at system level,” Dai said in an email interview.

At the same time, humanoids and advanced cobots are transitioning from experimental pilots to targeted deployment in manufacturing and logistics, he added.

This is driven by labour constraints and need for flexible automation, he said.

The evolution in robotics has been significant over the past decade, noted Ted Ng, HP’s senior manager of engineering and manufacturing.

Industrial automation then was largely about automating repetitive tasks, such as flipping, screwing, and performing the same motion repeatedly.

Robotics today is far more sophisticated, Ng said in an email.

“We moved from fixed automation to collaborative robots, or cobots, that can work alongside people, and we’ve seen the introduction of autonomous guided vehicles that can move materials around the factory with minimal human intervention,” he said.

There also are early developments in humanoid robotics, or “human bots”, but the technology here still is maturing, he added.

In Singapore’s manufacturing environment, where labour availability remains a challenge, these technologies could eventually help fill critical workforce and talent gaps, he noted, echoing Dai’s comments.

Armed with the relevant AI know-how, someone with limited skills also can potentially manage some systems on the manufacturing floor.

For example, AI can help them troubleshoot with natural-language prompts, cutting the time they otherwise will have to spend plowing through thick manuals, which can run into thousands of pages for industrial systems, said Vivekanand Patil, senior regional manager for robotics at Epson Southeast Asia.

They also may have to go back to the manufacturer or systems integrator before the problem can be resolved, Patil told FutureCIO.

With AI-powered prompts and data analysis, administrators can more quickly find solutions as well as better understand how to operate and manage the systems, he said.

This will prove beneficial for companies amidst the manpower crunch, he added.

AI can significantly cut the time needed for companies to figure out how to manage systems running in their factories, and allow them to more easily deploy automation and robotics in their manufacturing sites, Patil said.

Dai added that recent AI advancements also have provided the core intelligence layer that enables robots to perceive, reason, and act in dynamic environments.

In addition, global AI models, multimodal foundational models, and vision-language-action systems allow robots to interpret context and plan actions.

Dai said: “Generative AI (GenAI) accelerates training through synthetic data and imitation learning, while AI-native cloud platforms enable faster iteration and scalable deployment across fleets.”

Specific needs require customisable AI

Agentic AI can provide new capabilities that are better tailored to each organisation’s requirements, which is key for the sector, Ng said.

Manufacturing operates in a different environment, where the knowledge is highly specialised and lingo industry-specific, he explained.

Every production line also has its own requirements.

That means AI solutions often need to be tailored to the manufacturing setup within which they operate, he said.

Agentic tools can help optimise how production is sequenced and scheduled, he noted.

He added that HP is looking at the potential use of agentic AI in production planning.

For instance, a human planner currently has to decide the build sequence for multiple customer orders across the same manufacturing line, before handing it off to the shop floor where operators make further judgement calls.

HP is exploring whether agentic AI can help analyse these variables and recommend the most efficient production sequence, while balancing customer delivery timelines, quality requirements, and operational constraints.

For now, the vendor has made “meaningful progress” in adopting GenAI across its manufacturing floors, Ng said.

It developed its own internal manufacturing chatbot, called PISES Knowledge Management System, because there was no commercial solution available that understood its production environment and specialist terminologies, he said.

The PISES chatbot can answer technical queries and has multilingual capabilities, which are critical since HP’s manufacturing operations comprise workers from across the region, such as Myanmar and Nepal, he noted.

The chatbot supports six languages, including distinctions between Bahasa Malaysia and Bahasa Indonesia.

HP now hopes to leverage agentic AI to “elevate” its day-to-day manufacturing planning and operations, with production build plan optimisation as a leading use case.

Ng said: “The goal is to have AI agents dynamically manage sequencing across multiple customer orders on the same line, which is a workstream that is currently managed entirely by humans.

“We are still in testing, learning, and improvement stages, but we see significant potential and are actively testing where it can deliver meaningful value,” he said.

Finding more yield through AI transition

Singapore itself is aiming to increase its manufacturing sector’s value-add by 50% by 2030.

To do that, the nation should scale software-defined manufacturing and invest in simulation and digital twins, Dai said.

It also should look to accelerate workforce reskilling to support AI-driven automation, he said.

He added that Singapore should strengthen ecosystem partnerships and shared infrastructure for robotics training and deployment.

At the same time, it should avoid fragmented pilots and over-investment in immature humanoid use cases, focusing instead on scaling proven, high-impact applications across sectors, he said.

The Singapore government in February said it is working to transition the manufacturing sector to Industry 5.0, focusing on the collaboration between humans and AI-powered machines.

It wants to look at how technology can “augment human capability for better outcomes” and have robots working alongside humans at more workplaces, where humans will focus on quality control and problem-solving.

Asked how such goals could play out, Dai said: “Industry 5.0 will emerge through structured human-machine collaboration rather than full autonomy. Robots will handle repetitive and hazardous tasks, while humans focus on supervision and decision-making.

“Advances in multimodal AI and natural language interfaces will enable more intuitive interaction, supporting shared workflows where humans and AI-powered machines operate jointly in production environments,” he said.

Singapore has earmarked its Punggol Digital District as a testbed for robotics systems, which may be deployed on streets where they can be tested and assessed within a real-world environment.

The Infocomm Media Development Authority (IMDA) also is working with the Singapore Institute of Technology and various industry partners to determine the digital infrastructure required for “sustained and safe” robot deployment in live environments.

Safety risks that must be addressed

Organisations are concerned about high costs, deployment complexity, and uncertain returns, Dai said.

He explained that integrating AI into robotics often requires redesigning processes, upgrading infrastructure, and retraining workers.

Reliability and safety risks also are critical, since failures can cause operational disruption or physical harm, he said.

Furthermore, data requirements and model training overhead increase cost and complexity.

These push companies towards cautious, phased adoption strategies, the Forrester analyst said.

Security and added risks also are slowing down wider adoption, he said.

Enterprise customers are concerned about expanded attack surfaces, as robots integrate with cloud platforms and enterprise systems and this introduces risks of unauthorised control or disruption, Dai said.

Additional concerns include data exposure from sensor-rich environments, unpredictable AI behaviour in edge conditions, and limited transparency in decision-making, he said.

These factors, combined with unclear regulatory accountability, continue to slow large-scale deployment beyond controlled settings, he added.

Ng noted that manufacturing environments already are designed with security in mind, where there is no internet access on the production floor of most factories to protect operational data, intellectual property, and production processes.

“If your production know-how is connected to the open internet, you’ve created a vulnerability that can compromise your entire operation,” he said.

This underscores the importance of keeping operational data and AI workloads at the edge, rather than the cloud, in a manufacturing environment, he noted.

“The moment sensitive production data leaves the factory floor and travels to a cloud environment, you introduce latency, exposure, and potential points of failure that simply are not acceptable when intellectual property and operational continuity are at stake,” he said.

AI deployments must operate within tightly controlled environments and without cloud dependency, Ng said. He noted that HP’s own computing hardware, CPUs, GPUs, and laptops that are deployed across its manufacturing lines, operate as a standalone network.

Dai also urged organisations to apply zero-trust principles across IT and OT environments, segment networks, and enforce strict identity and access controls for both humans and machines.

“Continuous monitoring of robot behaviour, sensor data, and AI models is essential,” he stressed. “Firms should also use simulation to test edge cases and adversarial scenarios, and implement robust lifecycle management for models to ensure safe updates and rollback mechanisms.”

According to a Deloitte report, more than 500,000 industrial robots were deployed in 2024. This figure is projected to hit 700,000 by 2028, the consulting firm said.

It expects robots to be increasingly integrated with some form of physical AI, with 41% of business leaders expecting to transformational impact within three years.

Deloitte defines physical AI as the merger of physical systems with AI.

It predicts that physical AI integration will climb six-fold in two years.

“Physical AI marks the moment when intelligence moves off the screen and into the real world, transforming factories into learning systems that sense, decide, and improve continuously,” said Chris Lewin, Deloitte’s Asia-Pacific AI lead.

The report noted that there are barriers to adoption, such as cost and resource requirements and challenges in identifying use cases. It also pointed to gaps in talent and skills as a challenge, alongside data and technology availability.

Figuring out what goes where and when

Ng, too, pointed to technical features and expertise as key challenges HP faces in applying AI to its robotics systems.

“AI solutions for manufacturing cannot be bought off-the-shelf because domain knowledge is highly specialised and the operating environment is tightly controlled,” he explained.

Building and training in-house AI systems, such as PISES, also takes time and investment, he noted.

Another challenge HP faces is knowing when and where to deploy, he said.

“Not every manufacturing line warrants the same level of investment, and not every manufacturing line has the same needs. Therefore, we focus on areas where the benefits are clear,” Ng said. “Once a capability has demonstrated value, we explore how it can be applied more broadly.”