How can we enhance robot adaptability to better serve human collaborators?

Boston Dynamics' impressive robots, capable of jumping, dancing or opening doors, seem straight out of a science fiction film.

[Article issu de The Conversation écrit par Etienne Fournier-Aubret, Chercheur en Psychologie du Travail et Ergonomie, Université Grenoble Alpes (UGA) ]

Yet behind their physical prowess lies a more nuanced reality: although agile, these robots remain surprisingly limited in terms of intelligence. They carry out pre-programmed tasks, but still struggle to react autonomously to a changing environment.

A major current challenge in robotics is to integrate artificial intelligence systems into these machines, so that they become capable of adapting and making decisions, transforming them into true versatile assistants to humans.

But… why make intelligent robots?

The challenge is to design intelligent robots that interpret messages from the environment effectively to adapt to real life, filled with uncertainties (like sidewalks, imperfections in the ground) and variations (like the weather, a station with high or low traffic, etc.).

In industry, robots are particularly numerous compared to other sectors, but they are far from being adapted to changing environments, because they are limited by their perceptual capacities and by their ability to produce new behaviors for respond to unforeseen situations.

We think it would be interesting to design a human-robot collaboration based on understanding human capabilities — which the robot would complement.

The industry of the future calls for collaborative robots

By declaring the transition to Industry 5.0, the European Commission aims to improve the productivity of industries and the well-being at work of operators. This is why it proposes the use of collaborative robots (also called “cobots”) to assist humans in their work.

Indeed, if we note a growing number of robots installed in industry. In fact, there is a growth in the implementation of robots of 5% every year. However, these can have negative impacts on humans at work when technological development has pushed for innovation while ignoring the real needs arising from the workstation. To illustrate, a study showed that experts at their task develop frustration and have difficulty accepting technology that “helps” them when it is not adapted to their needs.

It now seems essential to focus the design of cobots on the real needs and constraints of humans who will have to collaborate with the robots at their workstation.

This implies understanding work “without the cobot”, then choosing with employees the tasks where the cobot could provide help. Adapting cobots to the needs of humans at work then takes two steps. The first requires adjusting the physical capabilities of the robot: its gestures (trajectory, speed) and the added tools (grippers, suction cups).

The second must allow the cobot to enter a reasoning loop close to that of humans: this loop begins with the perception of the environment, then the information taken from the real world is analyzed by the algorithm which directs the behavior of the cobot.

This is where artificial intelligence (AI) systems come into play.

How can we make robots more adaptable to the needs of their human collaborators?

Within the Marvin team at the Grenoble Computer Science Laboratory, we are developing AI techniques specially designed to take into account the physical environment of robots and allow them to reason and decide autonomously.

These artificial intelligence algorithms are based on the prior definition of the actions that can be carried out by the robot, the constraints to be respected for their execution, the state of the perceived world, as well as the target to be achieved.

The main difficulty lies in the large number of possible actions, even for seemingly simple problems. Sensors are used to obtain data on the cobot's environment in real time. With constraints such as “preserving human initiative” or “preventing humans from being exposed to dangerous materials” guiding robotic decision-making, algorithms allow the robot to continuously recalculate the “actions to be carried out” in taking into account developments in the real world and analyzing their consequences.

The experiments we are carrying out aim to analyze the impact of a cobotic collaboration (a cobot facing a human) during an industrial-type assembly task (the participants have a model and assembly parts that they must reproduce with the cobot as quickly as possible).

The results of our experiments show that human-robot collaboration at work can be generally beneficial to humans and their performance. Indeed, we show that the workload (physical and mental) of the human remains stable when carrying out the task with a collaborative robot, whereas this is not the case when he carries out the task alone or with another human. Interestingly, participants, whether or not they have worked with a cobot before in another context, have high trust in the cobot and expect the collaboration to be pleasant.

This type of device would also be an advantage for the company. Indeed, a human-cobot collaboration leads to a more successful task than an absence of collaboration or a human-human collaboration, the reliability of the work and its quality is increased in the case of a human-cobot collaboration. In our experiments, operators are also exposed to fewer risks (they touch fewer elements that are presented as dangerous) when the cobot adapts, thanks to AI, to the safety constraints imposed by the task.

On the other hand, there is a constant negative effect throughout all our experiments: the increase in the time to complete the task. Indeed, participants always take longer to complete the task when they collaborate with a cobot than when they collaborate with a human (or when they do not collaborate).

This effect is a little surprising, and suggests asking essential questions: could the development of robotic collaboration support efficiency at work? Our study seems to indicate that this might not be the case with current cobots, but that they would promote the quality of work and human well-being.

In conclusion, it is essential to design cobots equipped with AI which take into account the needs and constraints of humans and which reason using data from real observation of them. Thus, the design of cobots involves, above all, close collaboration between researchers in computer science, robotics and work psychology-ergonomics with the aim of jointly improving living conditions at work.