Robots deployed in shipyards and war zones are delivering patchy returns, as chaotic real-world conditions expose the gap between physical artificial intelligence (AI) hype and reality.
Practitioners’ estimates of how much deployed robotics is adding value range from about 15% to 50%.
“Soldiers, sailors and airmen will very quickly dispose of the robot if it’s not adding value to them and they can’t operate it quickly,” said Dan Anders-Brown, head of growth for advanced materials, electrification, autonomy and future systems at QinetiQ, a UK defense technology company.
“If it makes their life more complicated, they’ll fix the bayonet and they’ll do what they need to do rather than use some complicated system. Simplicity is key in my environment,” he said.
Defense robots face smoke, laser dazzle, obfuscation, bombs and people shooting at them on top of difficult terrain. QinetiQ has learned the cost of complexity through its robotics work in Ukraine.
Anders-Brown said only about half of the robotics deployed in defense works at the moment, with some adding no value at all. Autonomous vehicles running logistics resupply along a 2,000-kilometer road are among the success stories, as they take people out of vehicles and free them to go forward to the fight.
“You will find out very quickly how good your product is when you give it to a soldier, because they’ll either break it or they’ll find a different thing to do with it,” he said.
He said drone technology in Ukraine has advanced so far in three to five years that it has become a complete game changer, removing people from serious harm’s way.
Carlo Dentesano, head of robotics at the Digital Lab of Italian shipbuilder Fincantieri, said his company was at the beginning of the curve in getting returns from its robots. Asked to put a figure on it, he estimated 15% to 20%.
Fincantieri’s roadmap started with logistics robots, moved to quality monitoring and will eventually reach production tasks such as welding.
“Removing the man-hours is not how we intend to integrate robotics in our processes. We want the humans to collaborate alongside our robots,” he said.
He said Fincantieri wants its workers to build the skills to operate and collaborate with robots. Their knowledge of how the work is done would, in turn, help the machines become more efficient and adapt to edge cases that will always remain.
Robots in the nuclear sector are further from paying off.
Ertan Balaban, principal electronics engineer at the UK Atomic Energy Authority (UKAEA), said fusion is still a research effort rather than a reality, and robotics for the nuclear sector is itself a subject of research at a low technology readiness level.
“For the future fusion reactors, I can certainly say that robotics will be a big part of it. Most of the things have to be done by robots because of the hostility of the environment,” he said.
Pipe dreams of autonomy
The panelists spoke at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. Moderated by Christina Yan Zhang, chief executive of The Metaverse Institute, the session discussed how to keep robots running in variable environments and which hyped breakthroughs remain pipe dreams.
Dentesano doubts that one giant AI model will soon handle real factory work. He said robots should combine different ways of perceiving their environment rather than going directly from sensor to action.
“Having one single big VLA (vision-language-action) model like GPT style, based only on low-resolution images because they are heavy, will not deal with doing an actual job with the millimeter precision that we need in manufacturing. So you have to combine multiple systems,” he said.
“Things are changing and evolving very quickly, so I will be happy to be surprised. If we find out the ultimate model for AGI (artificial general intelligence) even in robotics, I will apply it immediately,” he said.
For Anders-Brown, the fantasy is senior commanders’ belief that robots will one day completely take soldiers out of harm’s way. He recalled seeing a drone in Kyiv a couple of weeks earlier that mistook a patch of grass for an obstacle, drove into a ditch and flipped over.
“Senior officers tend to think that some kind of advancement in AI is going to mean sentience. It’s just daft because they’re all being developed at about the same pace,” he said.
He compared it with the 1980s film WarGames, in which a computer concludes there is no point in fighting.
Balaban said fully autonomous robotic systems remain a distant prospect for super safety-critical settings, because the nuclear sector has zero tolerance for failure.
Fincantieri is studying flexible robot platforms, including humanoids, for its yards.
“You have to think about shipbuilding like metalworking and a construction site all together, a very chaotic environment. We have many phases of the process, many stages with different geometries and different environments,” Dentesano said.
He said designing a separate robot or mobile platform for each environment could leave the company with about 100 different robots within 10 years, each with its own maintenance, training and parts. Instead, it wants one platform flexible enough to work across all phases while keeping operations and maintenance scalable.
“We are mainly a team that acts as an integrator, so you have to find ways to solve complex problems and integrate different solutions from different vendors,” he said.
Start with the problem
Maria Clara Sayeg Ribeiro, who leads data and AI at the Centre for Advanced Manufacturing and Supply Chains of the World Economic Forum (WEF), said she would tell a plant manager with a budget for robots not to buy them just yet.
“The key question that we should ask is: we are acquiring robots and automation for what? What is the problem we are trying to solve? Is it productivity, quality or making the workplace safer?” she said.
She said the pace of new robots every year and new AI models every week has made companies forget to start with people. They should not go straight to the newest, shiniest model on the market.
She compared the gap between boardroom strategy and factory reality with Brazil’s 7-1 defeat by Germany at the 2014 World Cup, when the team had the best players and strategy. She said companies that fail to combine people, process and technology risk another 7-1 on the factory floor.
She said companies must then check that their factories have:
The right data
Sensors and perception
Interconnected machines
The right cybersecurity
The WEF’s Global Lighthouse Network has spent the past decade recognizing factories that escape “pilot purgatory” and scale up technology. Ribeiro said 94% of those lighthouses combine three technologies to reduce risk.
“It’s never only about physical AI or AI in isolation. It’s about combining and converging technology to make sure that you reduce this risk to the factory,” she said.
Anders-Brown said the UK’s Ministry of Defence doesn’t want to lead in every area of technology. It must lead where commercial industry has little presence, such as robotics for nuclear submarines, but elsewhere prefers to integrate ready-made commercial systems into complete solutions.
He said QinetiQ spends much of its time making commercial off-the-shelf systems work together, either as swarms or as independent systems. A recent one-way effector project fitted warheads and retrofit autonomy onto boats.
“The robotics will change. What we’re interested in is the ability to integrate new systems onto the sinew, the glue, the thing that holds it all together,” he said.
For fusion robots, the hurdle is the hardware itself.
“The problem with the current state of electronics is that it cannot withstand the high radiation level. This forces us to either explore new materials or find ways to protect our electronics,” Balaban said.
UKAEA works with universities on fundamental research into new materials and with startups on design techniques for radiation-hardened electronics. Where neither can do the job, it develops its own technology. For the past two years, it has been building a digital camera for its autonomous robots that can withstand one megagray of radiation.
The camera is designed to be integrated into UKAEA’s robotic systems, which will need to operate in the hostile environments of future fusion power plants.