Architecting Interaction: Towards a Human-Aware Cognitive Architecture

Traditionally, the fields of Cognitive Robotics and Social Robotics/HRI have evolved along parallel, often disconnected trajectories. Most cognitive architectures focus on creating autonomous agents capable of intelligent behavior in isolation, with interaction treated as an optional extension. However, insights from developmental psychology and neuroscience challenge this assumption. Human cognition is inherently social—shaped from the ground up by the need to live, act, and learn within a social group. Our perceptual, motor, memory, and decision-making systems are not just capable of social interaction; they are optimized for it. 

This perspective invites a fundamental shift in how we design cognitive architectures for robots: interaction is not an add-on, but a core mechanism for sense-making, prediction, and adaptation. Embodied communication, affective processing, internal state modeling are not merely features of HRI—they are cognitive necessities for agents meant to operate in human environments.

In this talk, we propose a principled framework for a human-aware cognitive architecture, where sociality is treated as a foundational constraint rather than a peripheral concern. By embracing interaction as a structural element of cognition, we aim to foster a higher degree of cognitive compatibility between humans and robots, ultimately enabling machines that are not just intelligent, but meaningfully integrated into our social and cognitive spaces.

Embodied Predictive World Models for Adaptive Cognitive Robotics

Contemporary robots often learn skills in narrow, pre-defined contexts, which limits their ability to adapt to new situations. Insights from embodied cognition suggest a different path: intelligence emerges not from abstract computation alone, but from continuous, sensorimotor engagement with the world. In this view, the body is not just an actuator—it is the medium through which perception, prediction, and reasoning co-develop. This talk will explore how predictive world models, inspired by the hierarchical and modular organization of the human neocortex, can integrate proprioceptive, visual, and tactile streams into unified, adaptive representations. Learned through intrinsic motivation rather than task-specific rewards, these models anticipate future states, adapt strategies, and reuse skills in unfamiliar contexts—demonstrating transfer, flexibility, and robustness. The discussion will highlight how embodied agents can autonomously acquire versatile repertoires of motor and cognitive behaviors—such as dexterous manipulation, coordinated gaze control, and sensorimotor synergy—without explicit supervision. By grounding learning in prediction and exploration, such systems develop richer internal models that are inherently generalizable. In this talk, we will examine the cognitive principles driving this approach, present empirical results with humanoid platforms, and outline broader implications for developmental robotics aiming to show how deeply embedding cognition in the body can yield robots that not only perform tasks, but continuously grow their capabilities.

Predictive Coding Light

Current machine learning systems consume vastly more energy than biological brains. Neuromorphic systems aim to overcome this difference by mimicking the brain’s information coding via discrete voltage spikes. However, it remains unclear how both artificial and natural networks of spiking neurons can learn energy-efficient information processing strategies. Here we propose Predictive Coding Light (PCL), a recurrent hierarchical spiking neural network for unsupervised representation learning. In contrast to previous predictive coding approaches, PCL does not transmit prediction errors to higher processing stages. Instead it suppresses the most predictable spikes and transmits a compressed representation of the input. Using only biologically plausible spike-timing based learning rules, PCL reproduces a wealth of findings on information processing in visual cortex and permits strong performance in downstream classification tasks. Overall, PCL offers a new approach to predictive coding and its implementation in natural and artificial spiking neural networks.