From early science fiction to factory floors, robots have captured the public’s imagination for decades. Yet despite promises of mechanical butlers and automated assistants, robotics remains surprisingly niche. Labour shortages, an ageing workforce, and AI breakthroughs are finally changing this reality.
Three stories define this shift in factory automation. From a warehouse where machines now correct their own mistakes to industrial sites where aging expertise is preserved as mechanical stand-ins take the risks once shouldered by people.
Finally, we will also explore the virtual realm where robots learn to walk long before their first steps on the factory floor, bringing complex tools within reach of those without advanced degrees.
These aren’t distant possibilities, they’re operational realities reshaping industries today, showing how humans and machines collaborate to maximise both efficiency and human potential.
How Ocado’s Robots Learned to Think for Themselves
Ocado is widely recognised for its colourful fleet of delivery vans, but the company’s true edge lies in its automated warehouses. At its 563,000 sq ft facility in Erith, North London, more than 3,000 robots zip across grids equal in size to three football pitches.
Coordinated by the DASH control system, they pick over 50 items every five minutes, a task that once took more than an hour on conveyor belts. The gains are tangible: food waste runs at just 0.5%, compared to an industry norm of 3 to 5%.
As Alex Harvey, Ocado’s chief of advanced technology, notes, “operationally grocery is very challenging… high cost to fulfil orders with short profit margins and a short shelf life.” To make the economics work, the company relies on machine learning to give its robots sharper perception and resilience.
Supervised models train bots to distinguish between visually similar products, smooth versus pulpy orange juice, or small versus large bags of salad. Harvey explains that these models “generate effective masks allowing the machine to identify the correct item to be picked thereby not picking up trash or incorrect items.” Accuracy at this level ensures customer orders remain reliable.
Beyond recognition, Ocado’s systems allow robots to learn from mistakes. Using behaviour cloning, bots mimic engineers who have logged more than 200 hours of remote picking and packing. The result is machines that self-correct.
“When it gets something wrong, it realises it has got something wrong, puts it back and recovers,” Harvey says. He adds that robots even adjust awkward items, such as large chocolate boxes, nudging them into bags with surprising dexterity.
To predict mechanical failures, Ocado repurposed Google DeepMind’s speech-to-text algorithm WaveNet to analyse robot movement patterns. Since speech is also a time-based signal, the team discovered that WaveNet could “characterise our speed profiles incredibly accurately, allowing us to detect bad moves and separate the good bots from the bad.”
Fighting the Brain Drain: Robots as Knowledge Preservers
Across industries with physically demanding environments, oil rigs, substations, gas plants, a critical problem is emerging. As baby boomers retire, decades of accumulated expertise walk out the door with them. Marc Dassler, CEO of Energy Robotics, has a solution: robots that don’t replace human knowledge but preserve and extend it.
“We’d like the baby boomer generation to stay in jobs as long as possible because they have a lot of knowledge, and they add a lot of value to our economy,” Dassler explains. His approach focuses on robots taking over the “dangerous, dirty, and dull” tasks while enabling experienced workers to apply their expertise from safer, more comfortable environments.
At Shell’s Energy and Chemicals Park Rheinland site in Germany, Energy Robotics’ ‘Drone-in-a-box’ solution has transformed tank farm inspections. Previously, workers climbed 25-metre-high tanks three times weekly, consuming more than 24 hours of a 40-hour work week just “running up tanks.”
Now, autonomous drones flutter over the tanks collecting the same inspection data using cameras and AI, while experienced operators review the results “from the comfort of an office,” as Dassler puts it.
Offshore facilities across Northern Europe (approximately 4,000) traditionally required expensive helicopter inspection trips. Now robots including Spot ‘dogs’ serve as remote eyes. These systems navigate using self-mapped digital twins, checking machinery multiple times daily and alerting operators with images when parameters fall outside predefined ranges.
For power utilities facing similar demographic pressures, robots equipped with sensitive gas detectors perform regular inspection rounds at substations and gas plants. At Bayernwerk’s Bavarian substation, robots inspect fence defects and detect overheating.”By digitalising their inspection processes with our automated inspection solution, Bayernwerk benefits from increased productivity, quality and safety,” Dassler notes.
The feedback from workers has been overwhelmingly positive. As Dassler concludes, “they’re really happy to have a robot tool doing the job which they don’t want to do. It’s just like having a dishwasher, because no one wants to do the washing up.”
Robots That Graduate from Simulation First
The robotics revolution has a PhD problem. “It should be that you don’t need a PhD before building a robot,” says Nvidia’s Amit Goel, director of product management for Edge AI and Robotics at intelligent computing giant Nvidia. His mission: democratising robot development for the masses.
The challenge intensifies as robotics expands beyond large-scale manufacturing into smaller businesses requiring “more variation and less volume.” Unlike automotive robotics handling repetitive tasks, today’s applications demand robots that are “easy to program, deploy, reprogram and scale.”
Nvidia’s solution centres on simulation technology enabling collaborative robot development. The Isaac Sim platform, built on Omniverse, allows globally distributed teams to work in simulated environments with high-fidelity physics. “There have been robot sims before but being able to collaborate within the simulation environment is game changing,” Goel explains.
Development begins with synthetic data generation through Nvidia’s Replicator platform. “Many industries are not equipped with sufficient sensors and do not have a lot of historic data to use for training the AI model,” Goel notes. Once trained in simulation, robots deploy with Jetson AI computers processing up to 275 trillion operations per second.
Amazon simulates warehouse designs and trains robot assistants before physical implementation, while PepsiCo optimises layouts in virtual environments before real-world investments. Continuous feedback loops stream data from deployed robots back to digital twins for ongoing retraining.
The economic implications are substantial: even small routing optimisations could save billions in the $9 trillion logistics industry. Industrial robot markets project doubling from 406,000 units in 2021 to 815,000 in 2030.
“We provide the tools so that your robot can live tens of years of life in a few days. That is how you accelerate development and make it more affordable,” Goel summarises.