Program
Program
The workshop will be held in conjunction with the 24th International Conference of the Italian Association for Artificial Intelligence (AI*IA 2026), on October 8th and 9th.
Embedded Files
8th October 2026
10:30 – Introduction to AIRO 2026
Session 1: Perception, Vision-Language Models & Semantic Reasoning
10:45 – Vincenzo Guarino, Emanuele Musumeci, Vincenzo Suriani, Daniele Nardi
iAm.md: Robot Skill Self-Assessment through Agentic Introspection for Unknown Open-Vocabulary Domains
11:00 – Alessandro Umbrico, Gianmarco Cirelli, Christian Tamantini, Francesca Cordella, Andrea Orlandini
Towards a Compositional Semantics of Object Parts for Open-World Robotic Manipulation
11:15 – Claudio Pelleriti, Valeria Seidita
From Detection to Disobedience: Semantic Command Arbitration with Open-Vocabulary Perception and Vision-Language Models
Session 2: Safety, Monitoring & Ethical Deliberation
11:30 – Gianni Randazzo, Valeria Seidita
Evaluating Ethical Deliberation as an Observable Process: A Knowledge Representation-Driven Approach
11:45 – Francesco Baldo, Luca Castri, Gloria Beraldo, Nicola Bellotto
Estimating the Causal Impact of Social Navigation Signals in Corridor Scenarios
12:00 – Maria Elena Griseta, Michele De Carolis, Giulio Mallardi, Filippo Lanubile
Statistical Multivariate Node-Level Anomaly Localization for Robot Runtime Monitoring
12:30 - 14:30 – AIxIA Awards & Lunch
14:30 Invited Speaker
Gabriele Costante
Department of Engineering, University of Perugia
Deep Reinforcement Learning in Vision-Based Mobile Robotics: Recent Advances and Causal Inference Strategies for Data Augmentation
Abstract:
Vision-based Deep Reinforcement Learning (DRL) has proven highly effective in learning perception-to-action mappings for mobile robots, turning high-dimensional visual inputs into control commands. This makes it a key enabler for autonomous systems addressing complex tasks in fields such as surveillance, agriculture, transportation, and human assistance.
This talk will present recent advances in vision-based DRL methods for both terrestrial and aerial robots, covering tasks such as active target tracking, quadrotor control, trajectory tracking, and collision avoidance. The methodological aspects and experimental results of the proposed solutions will be described, highlighting their benefits and limitations.
Drawing on these findings, the talk will then discuss the well-known issue of data efficiency in DRL, whose success strongly depends on the volume of training data. This is particularly critical when experience is collected on real platforms, where many essential interactions with the environment, such as those that would result in robot collisions, are restricted or prohibited.
To address this challenge, a novel approach will be introduced that leverages the causal inference paradigm to establish a counterfactual framework for generating experiences in the image-based offline training of DRL policies. By learning the Structural Causal Model (SCM) that represents the process dynamics, the method generates counterfactual samples for data augmentation.
Session 3: Task & Motion Planning, Affordances & Smart Manufacturing
15:10 – Alex Agliuzza, Matteo Baldoni, Cristina Baroglio, Roberto Micalizio, Consuelo Rubina Nava, Stefano Tedeschi
A Digital Twin-Based Testbed for MAS Applications in Smart Manufacturing
15:25 – Gloria Adegu, Alessandro Umbrico, Philippe Fillatreau, Thierry Louge, Stefano Borgo, Bernard Archimede
Investigating Semantic-Coupling of Task and Motion Planning
15:40 – Giuseppe Rauso, Riccardo Caccavale, Jaime Duque-Domingo, Alberto Finzi
Referential Ambiguity in Robot Task Planning over VLM-Generated Affordance Graphs
15:55 – Day 1 Closing Remarks
9th October 2026
09:00 – AIRO 2026 Day 2 Opening
Session 4: Reinforcement Learning & Autonomous Decision-Making
09:15 – Daniel Gigliotti, Flavio Maiorana, Luca Iocchi
Skill Composition for Legged Robot Reinforcement Learning
09:30 – Edoardo Zorzi, Flavio Maiorana, Alberto Castellini, Luca Iocchi, Alessandro Farinelli
Obstacle Avoidance in Hierarchical Reinforcement Learning
09:45 – Lucrezia Mosca, Maria Chiara Callari, Valeria Seidita
What does an autonomous robotic system need to know about itself? A symbolic self-model for deliberation and explainability information
10:00 – Davide Cordioli, Alessandro Farinelli, Alberto Castellini
Offline Learning of Shield Thresholds for Safe Robot Deployment
Roundtable Discussion
10:45 – Day 2 Closing Remarks
Page updated
Google Sites
Report abuse