Working on intelligent robotic manipulation (exploiting artificial intelligence techniques) opens up various important scenarios, ranging from prosthetics to industrial manufacturing. This is why an Italian-led EU project was born
The employment of artificial intelligence techniques in robotic manipulation it will allow robots engaged in multiple activities to make a significant leap in quality. The partners involved in IntelliMan, a Horizon Europe research project coordinated by the University of Bologna and which sees several other active Italian players, are convinced of this. The reason from which the project starts is allow a machine to carry out some of the most typical actions of man, such as those carried out by our hand, a rather complex limb with very large capacities for adaptation and interaction, able to exert a variable pressure on the manipulated objects according to the need. It’s a daunting task, especially if you plan to carry out these actions in a changing landscape.
AI has great potential for robotics, enabling a number of benefits in manufacturing. The main objective of employing AI techniques in robotics is to better manage the variability and unpredictability in the external environment, in real time or off-line. This was written by the international Federation of Robotics, which recalls the use of AI in robotics among the five key trends behind the growth of operational robots: worldwide there are 3.5 million units and it is a new record, as well as the value of the installations, estimated at 15.7 billion dollars.
Thus, the developments of the Italian-led EU project, financed by the European Commission with 4.5 million euros, take on reflections of strategic importance, both in the industrial field and in the welfare and prosthetic fields. In this last regard, it should be remembered that the University of Bologna will exploit its broad expertise in robotics and artificial intelligence to develop innovative solutions in the field of prosthetics, in collaboration with the INAIL Prosthetics Centre.
Takeaway
Intelligent robotic manipulation: the activity of the IntelliMan project
To better understand where the work of the teams involved in IntelliMan starts, the adoption and implications of artificial intelligence in robotic manipulation, we met Gianluca Palli, professor at the Department of Electrical Energy and Information Engineering “Guglielmo Marconi” of the University of Bologna and project coordinator.
«The use of artificial intelligence requires a large amount of data in order to be trained. In the handling field, however, this can represent a problem since there is a physical interaction with the environment and the need to guarantee high safety conditions. Furthermore, training must be efficient because it is necessary to make the most of the (few) data acquired. To achieve this result, what we intend to do – contrary to normal processes in which AI techniques are employed, in which there is a sort of “closed box model” whose content is unknown – is to create a predetermined structure for this learning activity, the result of which is obtained through the supervision of a tutor”.
In practice, a person takes care of guiding the manipulation system in the initial learning phase, starting from a predefined behavior, but with very limited performance.
«One of the salient aspects of the IntelliMan project is to employ a methodology based on shared autonomy: the trainer gives the robot the possibility of learning progressively, a bit like in the relationship between an adult and a child: the former leaves room for the latter to learn and gain confidence in carrying out a specific task or action».
This mechanism of shared autonomy ensures that the robot is able to achieve the assigned manipulation goal, requiring less input from the tutor who monitors the action of the machine and intervenes if necessary.
“It is a sort of teaching aimed at the progressive acquisition of autonomy, without violating the conditions of safety and without jeopardizing the technologies under study, with very high costs”, Palli points out.
The utility of artificial intelligence techniques in robotic manipulation
The use cases being studied by the IntelliMan project are four and concern: applications for prosthetics; two industrial use cases, for manufacturing and for the logistics of goods, in particular for agrifood; finally, a domestic service robot.
Speaking of the adoption of artificial intelligence in robotic manipulation, which techniques are used?
«Mainly, reinforcement learning is adopted, a machine learning technique for training based on the reward mechanism when the desired result is obtained, along the lines of the process used to train animals».
In the case of the machine, the operator rewards it when it performs a successful action by sending a signal; if not, he provides information on how he wants the operation to be modified to be effective. Another technique used is unsupervised learning, another subcategory of ML, in which models are adopted to verify the result obtained on the basis of a measurement system.
From prosthetics to the manufacturing industry: the many purposes of IntelliMan
The University of Bologna, coordinator of the artificial intelligence project in robotic manipulation, is active on two fronts: the prosthetic and industrial fields. In the first case, the researchers work on the possibility of develop embedded computing platforms, aimed at processing data and information acquired through wearable sensors, mainly used in the prosthetic field. For this purpose, the decades-long experience conducted by the research laboratory of the University of Bologna will be exploited for medical and prosthetic applications.
The other team, led by Professor Palli, focuses on robotic manipulation on a broad spectrum, from design to practical use. Together with the Slovenian ELVEZ, a manufacturer of specialized products for the automotive industry, electrical and mechanical engineering, it will handling systems for industrial production, in particular focusing on deformable objects: the agri-food sector is also strongly affected by these results. This could have important repercussions in a sector which constitutes the first item of the Italian manufacturing sector.
But the prospects that the project opens up are crucial for many areas:
«The aim is to create systems capable of carrying out actions by interacting with the environment in a safe, efficient and effective manner. There is a need for systems capable of carrying out activities in an intelligent way, being able to learn new tasks, from assistance and domestic robotics to industrial fields”.
The starting point, in the industrial field, are cobots: the goal is their evolution.
«Today, collaborative robots are able to carry out essentially the operations learned from humans in a repetitive way, with very limited adaptability. The perspective is to provide robots with cognitive capabilities, for interpreting the environment and for interacting with objects, adapting to the new scenario to effectively carry out the assigned task», concludes the project coordinator.