ICDL 2026 - 3rd Edition
at Kyoto University, Kyoto, Japan
Embedded Files
Call for Papers
In pursuit of fostering insightful conversation on the subject of learning, particularly at the intersection of Cognitive Architectures and Developmental AI, we invite participants to submit their contributions.
Beyond submitting papers presenting state-of-the-art research, we invite authors to share works in progress, reports featuring preliminary results, and critical reflections and position papers. This inclusive approach allows for exploring and identifying pertinent and novel discussion points that authors may wish to address during the ensuing panel discussions.
Important Dates
Important Dates
Submission Deadline: July 31, 2026 August 31, 2026
Notification of Acceptance: August 28*, 2026 September 07*, 2026
Camera-ready: September 4th, 2026 September 11, 2026
Workshop date and time: September 15th, 2026 - 13:45-17:05
Registration deadline: September 14, 2026 (https://attend.ieee.org/icdl-2026/attend/registration/)
*Submissions will be reviewed progressively as they are received, following a first-come-first-served basis, to guarantee enough time for early contributors to plan their trip to the conference.
Submission Procedure
Submission Procedure
Submission format: Extended abstract (2 to 4 pages) or a short position paper (1 page) - Page limit excluding references.
File format: PDF.
Templates: Standard RAS IEEE two-column paper templates, to be used for all submission types. Please refer to the available LaTeX or Word templates.
The submitted contribution must be written in English and does not need to be anonymized (single-blind review process). A panel of experts from relevant fields will be asked to review the contributions. Accepted papers will be invited for presentation either in oral or poster format, and the paper will be uploaded to the workshop website.
Please submit your contribution on the AI-CogDev forms.
For any questions, please contact us at aicogdev.workshop@gmail.com
Award
Award
This edition is endorsed by TC CO-RO, which will provide financial support for two awards recognizing early-career researchers, with a total value of $200.
Keynote Speakers
Matej Hoffmann
Matej Hoffmann
Czech Technical University
Understanding infant development through baby humanoid robots
Human newborns are at first virtually helpless, but within a year or two they learn to effectively control their bodies, to reach and to locomote. What brain algorithms are at the basis of this learning? Current knowledge on sensorimotor development is patchy. A common but unverified idea is that learning involves intrinsically motivated exploration of the body and the world, giving rise to coherent internal representations that subsequently allow effective motor control. In this talk, I will question these assumptions.
We longitudinally follow infants in three different scenarios - spontaneous behavior, reaching for tactile stimuli on the body, and reaching for objects presented visually - and use behavioral data from these three contexts to: (i) identify signatures of active exploration and goal-orientedness in spontaneous behavior, (ii) compare reaching to somatosensory and visual targets over development to understand the interplay of target localization and motor control.
To shed light on the mechanisms of the development of the "sensorimotor self", we instantiate these scenarios in baby humanoid robots. I will also show AI tools that we develop to extract infant 3D pose from videos, reattribute motion data to robots, and replay the motor, proprioceptive, visual, and tactile experience of infants.
Biography
Biography
Matej Hoffmann received the Ph.D. degree in Informatics from Artificial Intelligence Lab, University of Zurich, Zurich, Switzerland, in 2012. From 2013 to 2016, he conducted postdoctoral research with the iCub Facility of the Italian Institute of Technology, Genoa, Italy, supported by a Marie Curie Intra-European Fellowship. In 2017, he joined the Department of Cybernetics, Faculty of Electrical Engineering, Czech Technical University in Prague, where he is currently an Associate Professor and Head of the Humanoid and Cognitive Robotics Group. His research interests include humanoid and cognitive developmental robotics and development of the "sensorimotor self".
Francisco Cruz
Francisco Cruz
UNSW Sydney
Explainable Robotic Systems in Reinforcement Learning Scenarios
Robotic systems are more present in our society everyday. In human-robot environments, it is crucial that end-users may correctly understand their robotic team-partners, in order to collaboratively complete a task. To increase action understanding, users demand more explainability about the decisions by the robot in particular situations.
Recently, explainable robotic systems have emerged as an alternative focused not only on completing a task satisfactorily but also on justifying, in a human-like manner, the reasons that lead to making a decision. In reinforcement learning scenarios, a great effort has been focused on providing explanations using data-driven approaches, particularly from the visual input modality in deep learning-based systems. This talk focuses rather on the decision-making process of reinforcement learning agents performing a task in robotic scenarios.
Biography
Biography
Francisco received a bachelor's degree in engineering and a master's degree in computer engineering from the University of Santiago, Chile, in 2004 and 2006, respectively, and a PhD degree from the University of Hamburg, Germany, in 2017, working in developmental robotics focused on interactive reinforcement learning. In 2015, he was a Visiting Researcher at the Emergent Robotics Laboratory, Osaka University, and in 2018, a Visiting Researcher at the Polytechnic School, University of Pernambuco, Brazil. Francisco is a Senior Lecturer in Cognitive Robotics at UNSW Sydney since 2022, where he leads the Autonomous Agents and Robotics Research (A2R2) group. His current research interests include reinforcement learning, explainable artificial intelligence, human-robot interaction, artificial neural networks, and psychologically and bio-inspired models.
Vieri Giuliano Santucci
Vieri Giuliano Santucci
Institute of Cognitive Sciences and Technologies of the Italian National Research Council (ISTC-CNR, Rome)
Intrinsically Motivated Open-ended Learning in Autonomous Agents
Open-ended learning refers to the capacity of an agent to autonomously and continually acquire knowledge and skills, without relying on an exhaustive external specification of tasks or objectives. This capacity is central to the development of both biological and artificial agents able to operate in dynamic, unstructured, and potentially unknown environments, where the relevant competencies cannot be fully anticipated at design time. Drawing on developmental learning and cognitive science, intrinsic motivations offer a principled mechanism to support such processes: self-generated signals that drive agents to explore novel states, discover and select their own goals, and progressively learn associated skills, independently of any task-specific reward.
This talk examines the role of intrinsic motivations in autonomous open-ended learning. It introduces the notion of open-ended learning and its relevance for developmental robotics, discusses how different forms of intrinsic motivation support autonomous exploration, goal discovery and skill learning, and considers how these mechanisms can be implemented, also in relation to other learning approaches. Particular attention is given to the integration of motivational signals within cognitive architectures, where the management of motivations, the storage of acquired knowledge, and the learning of new behaviours must be coordinated and balanced. The talk aims to provide both a conceptual framing and concrete illustrations of how intrinsic motivations can be inserted into architectures for autonomous learning, and to highlight open challenges for the field.
Biography
Biography
Vieri Giuliano Santucci received his Ph.D. in Computer Science from the University of Plymouth, UK, following an M.S. in Theories and Techniques of Knowledge (Faculty of Philosophy, University of Rome "La Sapienza") and a B.Sc. in Philosophy (University of Pisa). He is currently a senior researcher at the Institute of Cognitive Sciences and Technologies of the Italian National Research Council (ISTC-CNR, Rome), where he leads the Autonomy Research in Intelligent Systems and Ethics (ARISE) group. His research mainly focuses on cognitive and autonomous robotics and reinforcement learning, developing architectures and algorithms for intrinsically motivated open-ended learning. He is also interested in related topics in cognitive science and computational neuroscience.
Contributions
Learning without predefined categories: a developmental robotic approach to audiovisual primitives
Learning without predefined categories: a developmental robotic approach to audiovisual primitives
Raphael D’Urso, Sofiane Boucenna, Alexandre Pitti
In the field of developmental robotics, if we want to model a robot’s learning process after that of a child, we must avoid the use of predefined categories. We designed a neural network capable of making cross-modal reconstructions using only visual and audio primitives. To do this, we used dual-channel theory, which processes audio and visual data simultaneously but independently to enable reconstructions. We trained our robot in a controlled environment so that it could learn to recognize up to 15 objects in an unsupervised manner. Our results indicate that the network spontaneously categorizes audiovisual signals into object classes. The audio-processing component of our network performs better, as it maintains perfect reconstructions for a larger number of objects than the visual component. However, both components are capable of making accurate reconstructions, as their accuracy remains above 80% even with 15 objects. For 38 unique objects in the COIL-100 dataset [1], we achieved accuracy rates above 90% for V2A reconstructions, demonstrating the potential to increase the number of objects. Furthermore, our model is able to recognize an object with an accuracy of over 50% for rotation angles of 10° or without rotation.
Download: Poster, Camera-Ready
Miminko: A Physics-Grounded Synthetic Infant Video Generator
Miminko: A Physics-Grounded Synthetic Infant Video Generator
Francisco M. López, Matej Hoffmann
Synthetic infant videos provide controllable and fully annotated data for developing methods to track and analyze infant movement. They are useful since accurate 2D and 3D ground truth is difficult to obtain from recordings of real infants. Existing simulated infant embodiments provide a complementary means of generating physically grounded behavior but are poorly suited for realistic video generation. To bridge these two approaches, we present Miminko, a physics-grounded synthetic infant video generator that bridges physical simulations with a realistic-looking infant body model. Miminko records the body state from arbitrary episodes of MIMo, a multimodal infant model, and retargets the resulting pose sequence to a realistic infant body mesh SMIL. The fitted infant is then rendered in Blender under configurable camera and visual conditions. Because the complete body and camera geometry remain known throughout the pipeline, our platform can export exact ground truth annotations: 2D and 3D landmark positions, depth maps, segmentation masks, and joint kinematics. Miminko connects physically grounded infant simulation and synthetic data generation for pose estimation and movement analysis.
Download: Poster, Camera-Ready
CogScore: Developmental Cognitive Benchmarks for Embodied Cognitive Architectures
CogScore: Developmental Cognitive Benchmarks for Embodied Cognitive Architectures
Leonardo de Lellis Rossi, Esther Luna Colombini, Ricardo Ribeiro Gudwin
Developmental cognitive architectures acquire cognitive and sensorimotor competences through interaction with the environment. Their evaluation requires experiments able to assess different abilities under controlled embodied conditions. In this paper, we present C OG S CORE, a framework for the multidimensional evaluation of cognitive architectures. The current implementation comprises four domains: sensing and perception, attention, motivation, and learning. Experiments can be executed through the online platform, where a compatible architecture is submitted and evaluated on the server, or locally, by installing C OG S CORE and submitting the generated results afterward. External architectures can be integrated by implementing the REST API used by the experiment controllers, and dummy models are available as implementation examples. We use CST/CONAIM-based agents in the experiments presented in this paper. The experiments show the use of COGSCORE for detecting sensory degradation, evaluating attentional and motivational behavior, and identifying developmental competence boundaries. The framework stores the task measurements and evaluation coverage associated with each experiment, allowing the results of compatible experiments to be analyzed and compared.
Download: Poster, Camera-Ready
Program
Venue
Kyoto University: Symposium Hall, the 5th floor of the west wing, International Science Innovation Building.
Campus Map: No. 69
Page updated
Google Sites
Report abuse