Robots have played a crucial role in industry for decades, with close to 4 million in operation today.² However, getting them off factory floors and into the real world has proven challenging. Despite our ability to build machines that easily surpass humans in terms of power and precision, imbuing them with the intelligence and perceptual capabilities to carry out even the simplest human jobs has proven tougher than expected.

Much of that is to do with the challenges of the physical environment. Traditionally, programming robots has involved creating detailed mathematical models of the robot and its environment, which are then used to deterministically control its actions. The approach is effective for well-defined tasks in predictable environments, such as assembling vehicles in an automotive factory. But this hand-engineered approach struggles with open-ended problems and the complexity encountered in environments that were not designed with robots in mind.

AI, and in particular machine learning, promises a solution to this challenge. These techniques learn how to behave by training on data, which means they are not constrained by the designer’s preconceptions. This makes them more flexible and adaptable. There has already been significant success in using these approaches to improve robots’ visual perception and navigational capabilities. But recent advances are now making it possible to apply AI to more complex functions like sensorimotor control, planning and human interaction.

This gives the potential for a new generation of mobile robots to have the physical, and even social, intelligence to work seamlessly alongside us. Some of the most promising uses cases are in areas such as logistics, manufacturing, agriculture, food preparation and care work. Thanks to investor exuberance following recent AI advances, funding for these possible futures has been growing significantly.³

That said, robots still struggle with tasks that humans find effortless, such as folding clothes or navigating crowded spaces. Solving these challenges with AI will require vast amounts of real-world robotics data that is both difficult and costly to obtain. In the face of shrinking workforces and ageing populations around the world, proponents say the investment will be worth it. However, it is also important to anticipate the potential disruption to employment that could ensue if a significant number of robots start entering the workforce.⁴

Key takeaways

Rapid advances in AI are transforming Robotic software, replacing hand-engineered, modular systems with massive models that can simultaneously solve perception, control, and planning. Caution is warranted though, as these models are data and power hungry, and their decision making is hard to decipher. To take full advantage of these new capabilities will also require breakthroughs in Robotic hardware. Better tactile sensing could allow robots to take on dextrous tasks currently out of reach for robots, for example. Advances in both battery technology and energy-efficient chips will be necessary to boost mobile robot run times. Open Data repositories are helping broaden access and training models in simulations could provide a potential workaround to the problem of AI’s insatiable thirst for data. Ensuring seamless Human-robot interaction will be more of a challenge. Large language models have opened up a promising new way to interface with robots, but imbuing them with social intelligence and the ability to predict human behaviours remains a distant goal.