We had an opportunity to interview a well known researcher at ISTC-CNR & Principal Investigator at PILLAR-Robots, Vieri Giuliano Santucci, about his work and goals for the PILLAR-Robots project.
- ISTC-CNR has recently hosted a successful 5-day meeting of the PILLAR-Robot Projects. Three full days were dedicated to the hackathon activities, with PAL Robotics’ TIAGo robot. How do you see the future developments and how this week-long meeting is gonna help the project achieve its objective of a fully autonomous robot?
The meeting in Rome was undoubtedly an important step for the advancement of the project. From a theoretical point of view, it enabled further alignment of the partners on the core topics of the project. In particular, several discussions and presentations focused on the deepening of the concept of purpose and its formalisation. Furthermore, the partners discussed how this construct is currently used and implemented in current models, and how it will be integrated in future ones. This activity made it possible both to identify how the concept of purpose influences the different modules of the cognitive architecture produced to date, and to steer a standardisation in its implementation, based on the formalisation produced within WP2.
From a more “technical” perspective, on the other hand, several activities took place during the meeting. The process of integrating the functions and algorithms developed within the different tasks was accelerated, benefiting from the daily interaction provided by the meeting. This process was supported by the comparison with the software structure prepared to manage the cognitive architecture: the partners began to transfer their code, adapting it to the ROS environment in which the system developed by the project will operate. Finally, thanks also to the presence of PAL technicians, real “hands-on” sessions were established during the meeting both to start working on the actual robot and to set up appropriate simulated scenarios in which to build the first versions of the use cases where the cognitive architecture will be tested.
- What’s the added value that ISTC-CNR is bringing to the consortium? And how do you see the project efforts helping those robotics stakeholders who are reading this article?
Within the Pillar Robot project, ISTC-CNR contribution is mainly related to our expertise in autonomous open-ended learning. Years of work on autonomous learning and intrinsic motivations are the primary components that my Institute brings to the project. Specifically, we focus on the role of intrinsic motivations and the capability of agents to develop autonomous and versatile learning.
The ISTC-CNR has integrated this expertise with the skills of other partners, particularly in the development of cognitive architectures for robotic control. Moreover, our machine learning (ML) expertise has also enabled ISTC-CNR to play a crucial role in formalising the concept of purpose and integrating it with more standard areas of ML, such as, for example, reinforcement learning.
Thanks to this activity, we are integrating open-ended learning—capable of adapting to new or unstructured scenarios—with the idea of purpose, thus directing versatile learning towards the aims of end-users. Such a new technology will significantly improve the possibility to employ artificial agents – and robots in particular – in complex and eventually unknown scenarios.
- Tell us a bit more about yourself, your organisation, and your role in the PILLAR project
My work over the years has mainly focused on developing autonomous systems and studying autonomy, both from the perspective of biological agents and artificial agents. My interest in the Pillar Robot project lies in the opportunity to continue research in this field and enhance the application of artificial agents in real-world scenarios, thereby expanding the usability of these systems.
ISTC-CNR covers various research areas, but regarding artificial intelligence, machine learning, and robotics, it concentrated on developing cognitive architectures for robotic control to improve the adaptability and versatility of artificial agents. As mentioned earlier, this is a crucial element of the PILLAR-Robots project, especially when combined with the concept of Purpose, which is at the core of the project itself.
This combination constitutes an innovation both in fundamental research on open-ended learning—introducing the necessity for alignment and biasing towards the needs and requests of developers or external users—and in the applied robotics field. Indeed, it allows us to leverage years of research in the field of autonomous open-ended learning, transferring it to applied systems that can be more effectively used in less structured environments where automation has so far been limited.