The PILLAR-Robots Perspective
During the project meeting held in Barcelona, the PILLAR project presented and advanced work across three complementary use cases: Industry, Edutainment and Agrifood. Together, these use cases show how the PILLAR architecture and technologies can support different robotics and AI scenarios, from industrial assistance and live robotic interaction to agrifood production and trustworthy AI validation.
Although each use case addresses a different context, they share a common objective: enabling robotic and AI systems to operate more effectively in real-world environments, where tasks, users, domains and operational requirements may change.
Industry Use Case: A Generalist Industrial Assistant
The Industry use case envisions the robot as a generalist industrial assistant. Its role is to perform repetitive and tedious tasks, such as tidying workspaces or bringing tools from across the workshop. By taking care of these routine activities, the robot can allow workers to focus on more specialised elements of their work.
In this context, open-ended grasping capabilities are essential. Industrial environments can involve many different objects, tools and manipulation tasks, meaning that the robot needs to adapt to varied situations rather than rely only on fixed behaviours. The ability to understand and carry out instructions phrased in open language is also crucial, as it allows workers to interact with the robot in a more flexible and practical way.
This use case therefore focuses on the robot’s capacity to support industrial workers through adaptable manipulation and instruction-following capabilities, contributing to more efficient and flexible workshop operations.
Edutainment Use Case: Adaptation with Real Robots
The Edutainment use case focused on demonstrating the adaptation capabilities of the PILLAR architecture in a live setup with real robots.
For this demonstration, realistic setups were implemented at PAL Robotics’ offices, including the Robobo and TIAGo robots, 3D cameras and control computers. The architecture ran in ROS 2 and was tested in practical conditions.
The demo showed that the PILLAR architecture was able to respond to user changes, referred to as purpose alignment, as well as to domain changes. This highlighted the flexibility of the architecture in interactive scenarios where the system must adapt to changing users, contexts and tasks.
By using real robots and realistic setups, the Edutainment use case demonstrated how the PILLAR architecture can support adaptable robotic behaviour beyond controlled or purely theoretical environments.
Agrifood Use Case: Industrial and Scientific Demonstrations
The Barcelona meeting also marked an important milestone for the Agrifood use case, with the definition of two complementary demonstrations. Together, they will validate both the industrial applicability of the proposed solutions and the scientific objectives of the PILLAR project.
The first demonstration adopts an industrial perspective and focuses on practical needs in the agrifood value chain. It will address three key tasks: sorting, identifying and removing unwanted or unsuitable objects, and packaging. These tasks will show how PILLAR technologies can support product classification, quality control, process safety and more efficient packaging operations.
The second demonstration focuses on the scientific dimension of the project. It will investigate how bias can affect an AI system’s ability to achieve its intended purpose. This supports PILLAR’s broader objective of developing AI systems that are not only effective, but also trustworthy, transparent and aligned with real-world requirements.
Together, the two Agrifood demonstrations connect practical industrial validation with research into trustworthy AI.
A Shared Focus on Adaptability and Trustworthy AI
Although the three use cases address different scenarios, they all contribute to the broader goals of the PILLAR project.
In the Industry use case, adaptability is shown through diverse grasp learning, cluttered manipulation and transfer to non-standard hardware. In the Edutainment use case, it is shown through live adaptation to user and domain changes using real robots. In the Agrifood use case, the planned demonstrations connect operational validation with the investigation of trustworthy AI, including the role of bias in achieving an intended purpose.
Together, the work presented and defined during the Barcelona meeting shows how PILLAR is advancing robotic and AI systems that are not only more flexible and capable, but also more aligned with real operational needs and responsible deployment.