DETROIT: Two years ago, a giant robot arm at a Stellantis assembly plant in Detroit kept freezing in place. The arm was programmed to pick panels off a rack and hoist them so they could be welded onto Dodge Durangos and Jeep Cherokees.
But whenever a panel sat even slightly out of position on the rack, the robot got confused and stopped working. The problem got worse as the racks aged. Sometimes the arm crashed entirely, taking up to an hour to reboot – valuable time on a manufacturer’s floor.
Such glitches are common, a symptom of a long-standing problem in industrial automation: Most robots can’t see. Pick-and-place robots are programmed to grab parts repeatedly from a specific place. If the part is even slightly out of position, the machine could damage it or break down.
But now, rapid advances in 3D cameras and artificial intelligence (AI) are making it possible for robots to recognise items and autonomously retrieve them, a development that could make factories more flexible in the future.
The Stellantis robot now has a special camera and computer that acts as its eyes and brain. The carmaker teamed up with Inbolt, a French startup that installed a purple box on the robot’s wrist, allowing it to recognise the panels it was supposed to pick up, pause to calculate their distance and position on the rack, and adjust its own actions accordingly.
“We watched this go from the bottleneck of the plant, to, ‘Boom!’” said Sean Woodall, senior vice president of assembly operations for passenger vehicles and SUVs at the plant. “We are back in business.”
Of the roughly 4.7 million industrial robots deployed worldwide, a vast majority “work blind,” according to Aaron Prather, director of market intelligence at the Association for Advancing Automation. Although some robot arms have sensors to help guide them, cameras can be far more effective. Vision technology and advances in AI have allowed robots to perceive the world around them, and perform a wide variety of tasks such as sorting through mixed items at a warehouse. Those are some of the fastest-growing applications in robotics, Prather said. “The current blind installed base will slowly be replaced,” he said.
But automakers like Stellantis have already invested in fleets of machines that can’t see. Stellantis’ managers were initially told they would have to upgrade their Fanuc robot arms to use the new AI technology. Buying the most up-to-date industrial robot with an embedded vision system can cost hundreds of thousands of dollars, so the managers at Stellantis looked for another way to upgrade their ageing machines, which had been making Dodge Durangos since 2010.
“One of the key things about Inbolt is, their system is able to work with old equipment,” said Loda Bazzi, who oversees a part of the Stellantis factory with nearly 1,000 robots. Her team wanted to compete with other automakers that were investing in AI.
Within weeks of installation, the arm was working smoothly. Employees who had previously been assigned to babysit the machine in case it had to be reset were reassigned. The Inbolt system worked so well that Bazzi had it installed on other robots.
The Inbolt system improved machine efficiency by reducing downtime, helping to make that plant a top performer in the company in less than two years, according to Bazzi. Leaders at the plant celebrated the achievement.
Inbolt declined to say how much its service cost Stellantis. And a spokesperson for Stellantis declined to estimate how much money Inbolt had saved the plant, but Joern Buss, a senior specialist in manufacturing at Arthur D. Little, a management consulting firm, said critical bottleneck failures at large auto plants could cost as much as US$100,000 (RM406,450) per minute.
Inbolt was founded in 2019 by graduate students at École Polytechnique and HEC Paris in France, who had to pitch a business idea for a class project.
Albane Dersy, who had a background in finance, was grouped with Louis Dumas, a software engineer, and Rudy Cohen, a researcher with a specialisation in applied mathematics. The three dreamed up a company that would put cameras on industrial tools to help human workers use them correctly, Dersy said.
They named it Inbolt and kept working on it after the class ended. The team won awards that gave them early-stage funding. One venture capital firm sent them to a business incubator in Shenzhen, China, where they saw how fast the field of robotics was advancing. They decided to change course and build cameras that could help robots instead of people.
Inbolt, which sells robot “eyes” as a subscription service, deployed its first system at a Stellantis plant in Hungary in 2024. Since then, it has been installed on more than 200 robots around the world. The company has raised roughly US$30mil (RM121.94mil) since its inception and employs 50 people around the world, including 15 in the United States.
“We have tripled our bookings year over year since our first deployment,” Dersy said. Companies hire Inbolt to get the most out of the machines they already have, she said, adding that machines with the system report up to 95% less downtime and 30% faster cycle times.
Bazzi, the Stellantis factory manager, selected an assembly station under relatively little pressure for the first Inbolt installation in Detroit: a stand-alone robot that installed hatchback-style rear doors. Her team worked closely with Inbolt’s to get the technology working. Once they were successful, Bazzi had it installed on the problem robot arm on the main assembly line.
It went so well that Bazzi decided to try it on other machines, including a robot that spreads sealant on car bodies with a nozzle. Before the camera was added, the nozzle frequently brushed against the vehicle and snapped off – as often as seven times in a nine-hour shift, forcing workers to spend 15 minutes replacing it each time. Now, it almost never breaks, Bazzi said, comparing its performance to “magic.”
Things haven’t always gone smoothly. Last summer, a cable connecting the camera to a quick-moving robot arm broke, knocking the camera offline. The robot stopped. Production ground to a halt. Bazzi placed an urgent call to Dersy – on Dersy’s wedding day.
“Don’t look at your phone,” Cohen, Inbolt’s CEO, told Dersy at the wedding dinner in the south of France. Dumas, the third co-founder, who is now chief technology officer, flew the next morning from the wedding to Detroit to fix it. – ©2026 The New York Times Company
This article originally appeared in The New York Times.
