(Tentative) Program

14:00 - 14:45: Exploiting Robot Abstractions in Episodic RL via Reward Shaping and Heuristics

Fabio Patrizi, Luca Iocchi (Sapienza University of Rome)

One major limitation to the applicability of Reinforcement Learning (RL) to many domains of practical relevance, in particular in robotic applications, is the large number of samples required to learn an optimal policy. To address this problem and improve learning efficiency, we consider a linear hierarchy of abstraction layers of the Markov Decision Process (MDP) underlying the target domain. Each layer is an MDP representing a coarser model of the one immediately below in the hierarchy. In this work, we propose novel techniques to automatically define Reward Shaping and Reward Heuristic functions that are based on the solution obtained at a higher level of abstraction and provide rewards to the finer (possibly the concrete) MDP at the lower level, thus inducing an exploration heuristic that can effectively guide the learning process in the more complex domain. In contrast with other works in Hierarchical RL, our technique imposes fewer requirements on the design of the abstract models and is tolerant to modeling errors, thus making the proposed approach practical. We formally analyze the relationship between the abstract models and the exploration heuristic induced in the lower-level domain, we prove that the method guarantees optimal convergence, and finally demonstrate its effectiveness experimentally in several complex robotic domains.

14:45 - 15:30: Multi-Agent Path Planning

Francesco Amigoni (Politecnico di Milano)

The lecture presents recent solutions proposed to address algorithmic problems that arise when multiple mobile agents need to plan coordinated movements within a common environment, for example, to avoid collisions. These problems are characteristic of automated warehouses, but also of video games and autonomous vehicles for passenger transport and agriculture. The lecture formally defines these problems and presents basic techniques for their resolution, discussing both their effectiveness (in terms of time required to complete missions) and their computational efficiency, highlighting the centralized or distributed nature of different solutions. Finally, some new extensions and variants of the basic techniques that leverage Reinforcement Learning methods are illustrated, along with practical application examples.

15:30 - 16:00: Coffee break + Poster Session

16:00 - 16:45: From natural language to robot task planning and execution

Luigi Palopoli, Marco Roveri (University of Trento)

We present a novel framework, called PLanning with Natural language for Task-Oriented Robots (PLANTOR), that integrates Large Language Models (LLMs) with Prolog-based knowledge management and planning for multi-robot tasks. The system employs a two-phase generation of a robot-oriented knowledge base, ensuring reusability and compositional reasoning, as well as a three-step planning procedure that handles temporal dependencies, resource constraints, and

parallel task execution via mixed-integer linear programming. The final plan is converted into a Behaviour Tree for direct use in ROS2. We tested the framework in multi-robot assembly tasks within a block world and an arch-building scenario. Results demonstrate that LLMs can produce accurate knowledge bases with modest human feedback, while Prolog guarantees formal correctness and explainability. This approach highlights the potential of LLM integration for advanced robotics tasks that require flexible, scalable, and human-understandable planning.

16:45 - 17:30: Learning human skills from egocentric videos: a path for human-level humanoids

Giuseppe Averta, Francesca Pistilli (Politecnico di Torino)

Notwithstanding decades of research, robots still struggle in performing daily living activities. Recent advancements on VLAs are pushing the limits of what current manipulators can do, but are still overly limited to short horizon tasks, and are not applicable to cases in which textual descriptions are not sufficient. In our research, we investigate alternative solutions which use videos of human activities collected from a first person perspective as a rich source of knowledge to i) capture the nuances of human planning for long horizon procedures, ii) provide a proper physical grounding, and iii) learn actionable policies for long horizon (procedural) tasks execution. To do this, we take inspiration from the inherent hierarchical structure of human cognitive processing, and foster the development of architectures that expose and highlight the hierarchical representations of human activities, which we can use to better understand (and replicate) human behaviour in daily living activities.

17:30 - 18:00: Poster Session