ABSTRACT: Can robots learn languages like humans do? This talk dives into Cognitive Robotics, where AI, Psychology, and Neuroscience join forces to teach robots how we think, learn, and communicate. Researchers use iCub and Pepper robots to explore how babies acquire words, gestures, and even abstract concepts like numbers. But it’s not just about words: robots also develop a Theory of Mind, learning to interpret human intentions and build trust. The discussion goes beyond tech, tackling ethics, AI’s limits, and the future of human-robot collaboration. Can machines truly understand us? Let’s find it out! BIO: Angelo Cangelosi is a Professor of Machine Learning and Robotics at the University of Manchester (UK) and co-director and founder of the Manchester Centre for Robotics and AI. Awarded a prestigious ERC Advanced Grant, his research interests are in cognitive and developmental robotics, human-robot interaction, trust, and AI for healthcare. With over £38M in research funding and 300+ publications, his work spans major international projects like eTALK, TAS Trust Node, and THRIVE++. He’s an editor-in-chief, conference chair, and author of the books Developmental Robotics and Cognitive Robotics (MIT Press). His research is shaping the future of AI and robotics.

1. Angelo Cangelosi
University of Manchester

2. Robotics and AI: Human-Centred
Companions for older people Cobots for Industry 4.0
Children in hospital Safety

3. What are the mechanisms and constraints
(biases, cues)
supporting cognitive development?
How is embodiment involved in
language acquisition?
The “Body as Cognitive Hub” Hypothesis
(Smith 2010; Morse et al. 2011)
Gavagai
Quine (1960)

4. “Merry-Go-Round” of
Amodal Symbol Systems
Example of self-referential, amodal network of word definitions in
Webster’s Dictionary (Roy 2005) Þ Chinese Room (Searle 1980)
Force: Energy or strength
Energy: Strength of force Strength: The power to resist force
Push: To press force-fully against to move
What’s the meaning of “Push” ?

5. Chinese Room Experiment
© Blink Films

6. Chinese Room Experiment
© Blink Films

7. Question: mita syoda tytto ?
• tytto: vuosi seitseman, vahan nalka, syoda muna
• tati: vuosi kahdeksankymmenta, paljon nalka, syoda kana
• kana : vuosi kaksi, paljon nalka, syoda siemen
• vuosi : kaksi , seitseman, kahdeksankymmenta
• nalka : paljon, vahan
• syoda : muna, kana, siemen
Dictionary
• kuinkamonta vuosi X ? è X vuosi A
• kuinka nalka X ? è X nalka B
• mita syoda X ? è X syoda C
Reply Rule Book
Chinese Room Experiment

8. tytto tati kana muna siemen
Dictionary
• tytto: seitseman vuosi , vahan nalka, syoda muna
• tati: kahdeksankymmenta vuosi , paljon nalka, syoda kana
• kana: kaksi vuosi , paljon nalka, syoda siemen
• vuosi: kaksi, seitseman, kahdeksankymmenta
• nalka: paljon, vahan
• syoda: muna, kana, siemen
Chinese Room Experiment
Finnish
paljon vahan

9. ü Children are slow, but efficient at learning a
language (vocabulary spurt) (Tomasello 2008)
ü Children use their body for situated interaction
(Smith & Samuelson 2010)
ü The brain integrates language and sensorimotor
knowledge (Pulvermueller 2005)
ü Children develop Theory of Mind (ToM) for
social interaction
Learning & Development
Robots can be easily pre-programmed to
memorise a dictionary, but cannot fully
understand the language they use

10. Cognitive
Developmental
Robotics
Child Psychology
Linguistics
Ethology Robotics
Neuroscience
Cognitive Psychology
Computer science

11. direct.mit.edu (open access)

12. Embodied Language Learning

13. Developmental Robotics
of Language Acquisition
• Cognitive architecture for developmental learning
– Experiments: first words, mutual exclusivity, U-learning,
– Collaboration with BabyLabs: Smith (Indiana), Horst
(Sussex), Floccia (Plymouth), Twomey (Manchester),
Marchetti (Cattolica Milan)

14. iCub’s Modi Experiment
Morse et al. (2015) PLoS ONE

15. Embodied Attention & Word Learning
Raggioli & Cangelosi (2022) ICDL

16. Epigenetic Robotics Architecture
AI
Architecture
ASR

17. Towards Learning
Abstract Words

19. Abstract Words: Numbers
• Skills development
– Finger counting
/Pointing
– Number Recitation
– Counting with/out finger
– Counting with/out pointing
Pecyna al. (2021)
Alibali & DiRusso & Pecyna et al.

20. ERC Advanced: eTALK
embodied Thought for Abstract Language Knowledge
TRAINING PILLARS
INTEGRATION DEMOS
WP4
MANAGEMENT, DISSEMINATION, IMPACT
WP5
FUNCTION WORDS
WP3
ABSTRACT WORDS
WP1
NUMBER WORDS
WP2
IMPACT
SCENARIOS
INTERDISCIPLINARITY
ROBOT PLATFORMS
Tutors for Maths
Companions for
Elderly

21. Trust in Human-Robot Interaction
Towards a Theory of Mind

22. Intention Reading
Vinanzi, Cangelosi & Goerick (2019, 2020, 2021)
• Cognitive architecture for intention reading from action and
social gaze
Which object is she going to grasp?
(Ambrosini, 2015)
• Intention (Mind)-reading is the
ability to understand the goal
pursued by someone through the
observation of physical clues
(mostly postural and gaze)
• Key factor in human survival and
basis for every other cognitive
ability; Developed with
experience (Woodward, 2009)

23. Intention Reading
Vinanzi, Cangelosi & Goerick (2019, 2020, 2021)

24. Trust in Human-Robot Interaction
Vinanzi, Cangelosi et al. (2018) Phil. Trans. Royal Society B

25. HRI for Trust and ToM
NB: Analysis show that there’s statistical
difference between the two groups.
Robot’s ToM Priming Results
• +ToM +Trust
• Other studies demonstrate role of social gaze (Zanatto et al. 2020)
Mou et al. (2020)

26. Take Home Message
• Developmental approaches
– Interdisciplinary approach
– Embodiment cues in development
– Trust and Theory of Mind
– Close match with empirical data
• Open challenges
– Open-ended learning and larger lexicons
– Neuromorphic robotics for brain-inspired models
– Explainable AI for Trustworthy Robots
– Robot companion and personal robotics
applications