Speaker: Angelo Cangelosi
Lecture Title 1: Understanding and Building Trust in Human-Robot Interaction: Mechanisms, Models, and Applications
Lecture Title 2: Neural Networks, Language Cognition, and Developmental Robotics: Bridging AI and Human Intelligence
Time: 14:00, April 29, 2026 (Wednesday); 9:00, April 30, 2026 (Thursday)
Location:290 Arts and Science Building
Abstract:
Lecture 1 will introduce interdisciplinary research on understanding and building trust in human-robot interaction, a field that integrates artificial intelligence, psychology, and social sciences to promote trustworthy human-machine collaboration. Using trust formation as a core case study, the lecture will demonstrate this research approach. Research in social psychology provides the theoretical foundation, revealing the critical roles of transparency and predictability in trust—a theory known as explainable AI (XAI). These findings directly inform robot interaction design and the establishment of human-robot trust. The research employs experimental robotics methods, with results demonstrated through experiments using Nao and Pepper robots. These studies reveal how communication mechanisms influence initial trust assessment and long-term cooperation. The same approach enables robots to learn complex concepts like social cues through behavioral consistency and emotional expression. The lecture will also introduce a novel computational model for studying trust dynamics and their relationship with collaboration. This work examines both the formation of human trust in robots and the artificial theory of mind used by robots to infer human trust levels. Results indicate that transparent decision-making and explainable robot behavior significantly enhance trust and collaboration efficiency. The discussion will extend to the application prospects of this method for social robots and service robots, while also addressing related ethical considerations, the ethical framework for trustworthy AI, and the current limitations of large language models in understanding social dynamics.
Lecture 2 will introduce interdisciplinary research combining neural network modeling, language cognitive mechanisms, and developmental robotics. This field aims to bridge the gap between artificial intelligence and human intelligence by exploring the computational nature of language acquisition and the developmental pathways of embodied intelligence. Using the neural basis of language evolution as a core case study, the lecture will demonstrate how computational modeling and robotic experiments are used to cross-validate theoretical hypotheses. Findings from cognitive neuroscience provide key insights, revealing the central role of predictive coding in language comprehension and production. This theoretical framework further advances the modeling of the hierarchical structure of language. The research employs a method combining deep reinforcement learning and developmental robotics, tested in simulation environments and on physical robot platforms (such as robot infants and desktop robotic arms). These experiments illustrate how multimodal sensory integration shapes the construction of early semantic networks and the induction of syntactic rules. The same framework enables robots to acquire conceptual metaphors and abstract relationships through active perception and social interaction. The lecture will also introduce a computational model simulating the process of language evolution, capable of reproducing the dynamics of lexical emergence and cultural transmission. This work simultaneously investigates the neural mechanisms of human language cognition and the pathways for machines to achieve explainable reasoning through neuro-symbolic systems. Research shows that introducing structured inductive biases and social learning mechanisms can significantly improve the generalization capability and robustness of robots' language skills. The discussion will further explore the application potential of this method in personalized educational assistance and neurorehabilitation robotics, while touching on related challenges of technological accessibility, ethical requirements for human alignment, and the fundamental limitations of current deep learning models in few-shot concept learning.
Personal Introduction:
Professor Angelo Cangelosi is currently a Professor of Machine Learning and Robotics at The University of Manchester (UK), and serves as the Co-Director and Founder of the Manchester Centre for Robotics and AI. He has successfully secured an Advanced Grant from the European Research Council (funded by UK Research and Innovation). His main research areas encompass cognitive and developmental robotics, neural networks, embodied language, human-robot interaction and trust mechanisms, as well as companion robots in health and social care. As project coordinator/principal investigator, he has led successful grant applications totaling over £40 million, including the EU FET project eTALK, the UKRI TAS Trust Node and CRADLE Prosperity Partnership, the US Air Force Research Laboratory project CASPER++, and numerous Horizon Europe and Marie Skłodowska-Curie Actions projects. Professor Angelo Cangelosi has published over 300 academic papers and books. He is currently the Editor-in-Chief of Interaction Studiesand IET Cognitive Computation and Systems, and served as Editor-in-Chief of IEEE Transactions on Autonomous Developmentin 2015. He has chaired several international academic conferences, including the ICANN 2022 in Bristol and ICDL 2021 in Beijing. His book Developmental Robotics: From Babies to Robots(MIT Press) was published in 2015 and has been translated into Chinese and Japanese. Cognitive Robotics, co-edited with Minoru Asada (MIT Press), is his latest publication in 2022.
[Editor: Yuhe Gao]