Envision a future where robots not only serve as invaluable assistants but also possess the remarkable ability to continuously learn and adapt to ever-evolving challenges with unprecedented skill and versatility. This captivating vision is the driving force behind the pioneering European initiative known as PILLAR-Robots (Purposeful Intrinsically motivated Lifelong Learning Autonomous Robots). This ambitious project is dedicated to forging a revolutionary generation of autonomous agents, distinguished by their unmatched capabilities and the innate motivation to thrive in an ever-changing world.
At its core, PILLAR-Robots is all about developing robots that can thrive in the unknown. Picture a robot that can step into unfamiliar scenarios, adapt to different environments, and display versatile behaviour, much like a human. These robots are designed to handle unstructured and ever-changing situations with ease leveraging the power of intrinsically-motivated open-ended learning framework. Intrinsically-motivated open-ended learning (IMOL) refers to a learning process in which an entity, such as a robot or an AI system, is driven to acquire knowledge and skills driven by its inherent curiosity, interests, or internal motivations, rather than external rewards or directives. It involves continuous learning without a predetermined endpoint, allowing the entity to explore and adapt to a wide range of novel challenges and scenarios autonomously.
At the same time, these robots need to follow human prescriptions, bounding their autonomy towards the “purpose” designers and end-users have decided for them. To achieve these goals, PILLAR-Robots has divided its work into several Work Packages (WP), one of which is WP3 that focuses on the critical aspect of creating a module that provides motivation to the robots. In particular, WP3 aims to integrate intrinsic motivations (guiding autonomous learning) and purposes (incorporating designers and end-users requests).
A first task of WP3 is to create a smart attention module. This module acts as a filter, allowing robots to focus on elements of the environment that align with their internal motivations, making its information-gathering activity purpose-driven and efficient. Moreover, WP3 will allow robots to discover new goals autonomously. Imagine a robot exploring its environment and, like a curious explorer, coming up with new objectives based on both its internal desire for novelty and curiosity and the requests provided by its designers and users. In essence, it’s about striking a balance between autonomy and guidance. But perhaps the most significant achievement of WP3 is the development of a motivational engine. This engine is like the robot’s internal source of inspiration, driving it to learn, adapt, and serve human needs effectively. It takes all the different motivations necessary for autonomous learning and adaptability and weaves them into the robot’s cognitive architecture. This means that the robot can dynamically balance its motivations, ensuring that it behaves both versatilely and purposefully.Finally, WP3 will test this motivational engine in real robotic scenarios to prove its efficiency. The success of WP3 will pave the way for integrating this module with others developed in different parts of the PILLAR-Robots project into a unique and integrated hierarchical robotic architecture.