Program
Workshop Overview
Topic 1: High-Level Learning, Task-Level Learning, and Deliberation
Francisco Martín Rico, Rey Juan Carlos University, ES
Title: Plan Replanning and Repair at PlanSys2
Abstract: PlanSys2 is the reference planning framework in ROS 2. It combines classical planning with Behavior Trees to execute plans on real robots. However, replanning in such frameworks is typically limited to canceling the current plan and generating a new one from scratch. In this talk, we will explore alternative approaches for adapting to changes in conditions or goals, focusing on repairing the current plan to align with the new situation. We will discuss how various strategies, including the use of Large Language Models (LLMs) to assist in this task, can enhance the efficiency of plan execution.
Richard J. Duro, Universidad Coruña, ES
Title: The Purpose Framework: A Path to Useful Autonomous Robots
Abstract: This talk will introduce the main issues that are being addressed in the PILLAR-Robots project related to Lifelong Open-ended Learning Autonomy (LOLA). PILLAR-Robots seeks to create the basis for developing a new generation of robots endowed with a higher level of autonomy, able to determine their own goals and strategies creatively building on the experience acquired during their lifetime to fulfil the desires of their designers/users in real-life application use-cases. The concept of Purpose, drawn from the cognitive sciences, is operationalized to increase the autonomy and domain independence of robots during autonomous learning, leading them to acquire knowledge and skills that are actually relevant for operating in target real applications. In particular, the talk will describe the purpose framework that has been developed as an integrated motivational structure that supports algorithms for the acquisition of purpose by the robot, ways to bias the perceptual, motivational and decision systems of the robots’ cognitive architectures towards purposes, and strategies for learning representations, skills and models that allow the execution of purpose-related deliberative and reactive decisions. These ideas are being applied in three different application fields characterized by different types and levels of variability: Agri-food, Edutainment, and unstructured Industrial/retail.
Christian Henkel, Bosch Research, DE
Title: How we tell the robot what to do? Deliberation in CONVINCE
Abstract: In this workshop talk, we explore the question: How do we tell a robot what to do? focusing on deliberation within the CONVINCE project. Deliberation, also known as task-planning or mission-planning, is essential for enabling robots to operate autonomously and adaptively in dynamic environments.
We will present the tools and approaches developed in CONVINCE, including formal methods, machine learning, and anomaly detection, and discuss how these support robust decision-making. Drawing from insights gained through the Deliberation Working Group and a ROSCon workshop, we will address key open questions:
What do users need to enable effective robot autonomy? How does deliberation differ from traditional industrial robot programming? How can we balance low-code solutions with flexible configuration? How do we ensure explainability and interpretability, aligning with Industry 5.0 principles? This talk aims to engage the audience in a discussion on the unique challenges of modern deliberation, fostering ideas for bridging the gap between robot programming and autonomous behavior.
Topic 2: Skill-Level Learning and Manipulation
Néstor García, Technology Centre of Catalonia Eurecat, ES
Title: TBD
Dimitrios Giakoumis, CERTH, GR
Title: Advancing the Physical Intelligence and Performance of Robots Towards Human-Like Object Manipulation
João Silvério, German Aerospace Center, DE
Title: Improving skill learning and generalization via deliberate human inputs
Abstract: When acquiring new manipulation skills, robots must navigate uncertainty while leveraging interactive guidance to adapt to dynamic, real-world environments. By employing uncertainty-aware motion primitives to model and refine skills and constrained reinforcement learning to explore within safe and practical limits, robots can unify human demonstrations, corrective feedback, and task constraints. In this talk, I will present recent results on uncertainty-aware, human-centered frameworks that facilitate adaptive and reliable robot learning in manipulation tasks.