AI-generated summary
The fusion of artificial intelligence (AI) and robotics aims to create machines that can adapt swiftly and effectively to unpredictable environments. Despite progress, AI applications in robotics remain nascent, with collaborative robots (cobots) being the most notable advancement. Cobots are designed to work safely alongside humans, accounting for about 10% of robot installations worldwide in 2022. However, AI still struggles with real-world physical interactions that require understanding complex cause-and-effect relationships and manipulating objects as seamlessly as humans can. Current research focuses on enhancing robots’ semantic intelligence—enabling them to comprehend context and human behavior to make better decisions. Techniques like imitation learning, where robots learn by observing human actions, have led to robots capable of tasks such as peeling vegetables or cooking, demonstrating AI’s potential to improve robot versatility without explicit programming.
Service robotics is evolving to offer more natural, empathetic assistance, especially for the elderly, with robots like Pepper and ElliQ providing companionship and engagement. Mobile robots trained through virtual simulations can adapt to new environments, as seen in the humanoid robot Digit. Emerging technologies like soft robotics mimic biological systems for delicate tasks in medicine and social care. Urban applications include autonomous vehicles interacting with smart infrastructure and delivery robots reducing emissions. Brain-computer interfaces could soon enable control of devices through thought, enhancing accessibility. Despite these advances, challenges persist around data privacy, liability, and workforce impacts, underscoring the need for careful regulation and reskilling efforts as AI-driven robotics continues to expand.
The merging of these two technological fields is revolutionizing our relationship with machines, making them not only more efficient but also more "human"
The main objective of merging artificial intelligence and robotics is to make machines able to adapt, in real time and more easily, to the variability and unpredictability of the environments in which they operate. However, in the report Artificial Intelligence in Robotics, by the International Federation of Robotics (IFR) in Frankfurt, it is mentioned that “artificial intelligence is still in its infancy in terms of applications in the field of robotics”.
According to the IFR document, the most significant development trend of this integration with AI is collaborative robotics: the so-called cobots, intended to operate safely alongside humans, thanks to their sophisticated mechanics and control electronics. Collaborative robotics applications, the report states, reached 10% of annual installations worldwide in 2022 (latest available data).
The problem is that when AI is confronted with physical reality, the boundaries are still significant. Unprogrammed interaction with an unstructured environment requires the ability to give meaning to the objects the robot manipulates and the actions it performs, to understand the cause-and-effect relationships between the actions performed and the outcome they produce in the environment, advanced perceptual capabilities, and increasingly precise manipulation.
For humans, operations such as taking a package of pasta from a shelf, after visually recognizing it, moving other objects (jars, cans, etc.) that get in the way, are tasks that are within the reach of a child, but in many cases still exceed the current capabilities of robots.
How to make robots learn
Making the robot understand the meaning of human actions to predict its next moves (if a person picks up a screwdriver, it is possible to presume what their next action will be) and act accordingly, is something that is being perfected in robotics labs, but is not yet ready to be transferred to industry.
For this reason, research working on the application of artificial intelligence in robotics, in addition to improving the understanding of voice commands, is currently focused mainly on semantic intelligence. It is about getting the robot to understand the context and the person with whom it is interacting and, based on this, make the appropriate decisions.
On the other hand, advances in artificial intelligence are accelerating a change that had already begun. AI algorithms allow robots to analyze data, recognize patterns, and learn from experience, continuously improving their skills. One of the most widely used methods is imitation learning, in which robots observe and replicate human actions. For example, the Toyota Research Institute, in collaboration with MIT and Columbia University, has developed robots capable of peeling vegetables or pouring liquids thanks to human demonstrations. These technologies allow machines to learn new skills without detailed instructions, making them more versatile and adaptable.
In a similar demonstration, a Stanford team used a relatively inexpensive commercial robot, costing $32,000, to perform complex handling tasks such as cooking shrimp and cleaning spots. The robot, called Mobile ALOHA quickly learned these new skills thanks to artificial intelligence. It only needed 20 human demonstrations and data from other tasks, such as tearing a paper napkin or a piece of duct tape. Thus, the researchers found that AI can help robots acquire transferable skills: training in a certain task can improve performance in others.
Assisting humans naturally
The ability to learn from different types of inputs and generate appropriate responses is the basis for advances in service robotics, which aims to create machines capable of assisting people autonomously and proactively. For example, robots designed for elderly care can understand people’s tone of voice and body language, responding empathetically and appropriately.
This makes a more natural and personalized interaction possible, thus improving the quality of life of those who are in situations of isolation. This is the case of the famous anthropomorphic robot Pepper, already used in some nursing homes to provide company to the elderly, or the robots developed by startups such as ElliQ, which use AI to converse, play games and encourage physical and mental activity in the elderly.
Another rapidly growing sector is mobile robotics, with models using sim-to-real reinforcement learning techniques to train in virtual environments before performing real-world tasks. This methodology allows robots to adapt to new contexts without needing detailed programming for each scenario. This is how the humanoid robot Digit works, the result of a collaboration between the universities of Oregon and California, Berkeley, whose advances were published in Science Robotics. Digit now knows how to walk and carry objects even in variable environments, never before known.
A sea of possibilities
The use of soft robotics could be another very interesting innovation in combination with AI. This technology is inspired by biology and aims to create robots with structures and movements similar to those of living organisms. They can be used in sectors such as medicine to perform minimally invasive interventions, or in social robotics, where they can interact more naturally with humans.
Urban infrastructure is also being experimented with where autonomous vehicles can interact with traffic lights and smart roads to improve traffic safety and efficiency. As well as delivery robots that could be used to transport goods in cities, reducing costs and emissions. And what once seemed like science fiction could soon become reality, namely the use of brain interfaces to allow people to control devices and interact with technology using only the power of thought — a technology that could break down many barriers for people with disabilities.
Despite the obvious advances in AI applied to robotics, there are still significant challenges to overcome. Data collection and management, for example, raises privacy and security concerns, necessitating rigorous measures by companies. The question of liability is also crucial: in the event of accidents caused by robots, clear regulations are needed to define who is liable. In addition, automation could lead to job losses, but reskilling programs can help people join new roles. The challenges are many, the opportunities even more so.
