Skip to main content

Predictive processes for intelligent behaviours, sensory enhancement and subjective experiences in robots

Periodic Reporting for period 1 - Predictive Robots (Predictive processes for intelligent behaviours, sensory enhancement and subjective experiences in robots)

Reporting period: 2019-05-15 to 2021-05-14

Our society is increasingly in need of smarter and adaptive artificial systems that can support us in different tasks. However, programming robots to work in dynamic human environments is costly and challenging. Out of laboratories, their functioning mostly relies on pre-engineered behaviours and the way they perceive things is strongly biased by their algorithmic design. This limits their adaptivity and efficacy. On the contrary, humans incrementally acquire capabilities, and refine and adapt them over time. Experience and factors such as emotions and moods shape the way we learn and further perceive ourselves and the world around us.

This project studies developmental processes with the aim of producing more adaptive robots. It researches learning mechanisms, exploration behaviours and artificial curiosity to let robots autonomously acquire knowledge about their bodily capabilities. Experience can teach robots, like infants, what to expect from further interactions with the surroundings. This research studies the formation of predictive capabilities, i.e the capability to anticipate upcoming sensory information. Current theories on brain functioning consider this as a fundamental computational component for motor control, cognitive development and adaptivity in humans.

In particular, this project had three objectives:
O1. implementing developmental mechanisms for the acquisition of predictive capabilities in robots. Mechanisms for curiosity-driven exploratory behaviours and continual learning, based on deep neural networks and memory consolidation techniques and in the context of multi-modal and high-dimensional sensory spaces, have been developed. These have allowed different robotic platforms to incrementally form predictive internal models.
O2. Enhancing robot perception through predictive processes. Anticipating sensory observations can allow modulating the focus of attention towards the most relevant part of the input for the task at hand. For instance, expected or non-relevant portions of the visual input can be filtered out, leaving thus the novel or most informative part of the scene in the field of view. Moreover, self-regulatory mechanisms can be enabled by monitoring the prediction performance during learning, e.g. modulating the allocation of computational resources or arbitrating internal goals, depending on the novelty of incoming sensory data.
O3. Studying predictive processes as a prerequisite for an artificial self. We perceive and anticipate the results of our own actions differently than those of others. The mismatch between expectations and observations - i.e. the prediction error - has been argued to be a marker characterising the experience of the “self”. This project has studied the role of prediction errors and their dynamics in robot self-perception. Moreover, it produced interdisciplinary theoretical work on the topic of self-perception in humans and robots.

Ultimately, predictive processes can represent a low-cost strategy to enhance current sensing technologies, and studying them can also help shedding light on phenomena such as (self-)perception in humans and on their possibility in artificial systems.
This project investigated the development of predictive internal models in robots. Behaviours driven by artificial curiosity enabled robots to explore their surroundings and to learn about their bodily capabilities. Intrinsic motivation algorithms for high-dimensional goal spaces have been developed in Work Package 1. Incremental learning techniques in deep neural networks have been studied. In particular, solutions using prediction-error driven memory consolidation in deep neural networks to prevent catastrophic forgetting issues have been implemented. Experiments have been carried out using sensory observations from different modalities, including visual, proprioceptive, auditory and motor, on different robots, such as humanoids and agricultural platforms.
This project studied multi-modal integration of sensory data and predictions, and its dynamic modulation during learning. Attenuating sensory predictions has been studied as a mechanism to allow perceptual enhancement in artificial systems (Work Package 2). The role of predictive processes in self-perception and subjective experience has been the focus of Work Package 3. First, sensorimotor datasets from simulated robots have been acquired. Experiments on prediction error analysis and self-regulatory mechanisms based on prediction error dynamics have been carried out. Moreover, theoretical work on related topics has been produced through interdisciplinary collaborations.

The project produced 13 scientific papers published in international journals and conference proceedings, and two datasets, as listed in the project website. Many of such works resulted from interdisciplinary collaborations with cognitive scientists, developmental psychologists, philosophers, media scientists. Collaborations also produced successful research grant applications. Project results have been disseminated through a variety of activities and in different venues, including social networks, conferences, workshops and fairs. A special issue in IEEE Transactions on Cognitive and Developmental Systems has been organised on the topic of “prediction and perception in humans and robots”, and a workshop has been held at the host institution and on the topic of “predictive processes for motor and cognitive development in robots”. An animated cartoon describing the topics of the project and aimed at reaching a general public has been produced (
The project has been associated with the German DFG Priority Programme “The Active Self” and with the DFG project on “Adaptive Architectures for the Transferability of Greenhouse Models”. It participated in the pilot program of the EU-H2020 OCRE project (Open Clouds for Research Environments), which provided funds for running deep learning experiments on cloud computing resources.
The following progress beyond the state of the art has been achieved:
- a novel intrinsic motivation architecture for the incremental acquisition of predictive capabilities
- novel self-regulatory mechanisms for robots based on prediction error monitoring
- new strategies for stable continual learning based on prediction-error driven memory consolidation
- new strategies for enhancing robot perception through sensory attenuation and dynamic modulating of multi-modal integration
- advances in the study of self-perception in robots and in environment perception (e.g. echolocation)
Two novel multi-modal datasets for studying sensorimotor learning in robots have been also produced.

State-of-the-art has been advanced also with interdisciplinary theoretical work on the "minimal self" and on predictive processing. A nomination for a Best Paper Award in IEEE ICDL/Epirob has been received.

Overall, the research on self-monitoring processes, sensory enhancement and self-perception represents a step forwards in the implementation of more adaptive and autonomous artificial systems, thus widening market potential for robotics and its socio-economic impact. This project initiated a collaboration with agricultural scientists in the innovative greenhouse sector, by implementing adaptive models for facilitating the transfer of knowledge between laboratories and production facilities, potentially supporting crop yield increase and cost reduction.
Predictive Robots logo