Closed-loop robotic systems powered by spiking neural networks for learning sensorimotor control in autonomous operation.

Neural networks for representation, computation, and learning in sensor data processing, integration, de-noising and decision-making.

Building a foundational knowledge base by applying antifragile system design, analysis, and development across domains. We are interested in formalizing principles and an apparatus that turns the basic concept of antifragility into a tool for designing and building closed-loop systems that behave beyond robust in the face of uncertainty.

Research in the field of Tiny Machine Learning (TinyML) explores the optimization and execution of AI-based processing chains on embedded computers, focusing on low-power neuromorphic sensors and computers.

Efficient sensorimotor processing is inherently driven by physical real-world constraints that an acting agent faces in its environment. Sensory streams contain certain statistical dependencies determined by the structure of the world, which impose constraints on a system’s sensorimotor affordances. This limits the number of possible sensory information patterns and plausible motor actions. Learning mechanisms allow the system to extract the underlying correlations in sensorimotor streams. This research direction focuses on the exploration of sensorimotor learning paradigms for embedding adaptive behaviours in robotic systems and demonstrating flexible control systems using neuromorphic hardware and neural-based adaptive control.

Traffic simulation is a crucial tool for testing and evaluating hypotheses in both the design of autonomous driving systems, where modelling and calibrating driver behaviour is key. In this research, we introduce a novel approach to calibrating driver models by combining fuzzy logic and inference to produce a plausible parametrization that reproduces drivers’ behaviour peculiarities and road traffic characteristics. The calibration system uses solely vehicle trajectories that describe cars’ movements through the road network, with a resolution of time and space enough to allow for determining the lane of the road used, speed, and acceleration at every second of the survey period.

Predictive maintenance of industrial automation systems to reduce unplanned stops and maximise equipment life cycle. “If component A is showing over 20% baseline vibrations, while the temperature rises 0.5 degrees in component B, and the noise level rises with 10.6dB then it is likely that the doors will break in about 5 to 7 days.” The project aims to learn normal operative parameters and sensory correlations to: 1) define bounds for min/max ranges of critical parameters and 2) generate alarms for anomaly situations. We research neural network streaming algorithms capable of explaining the variance-covariance structure of a set of variables in a stream through linear combinations. The essential neural algorithm is leveraged by novel incremental computation methods and systems operating on data streams and is capable of achieving low latency and high throughput when learning from data streams while maintaining resource usage guarantees for predictive maintenance tasks.

Road traffic congestion poses serious challenges to urban infrastructures. Such dynamical loading of the vehicular arteries impacts both the social and economic lives of people. The main goal of this work is to propose a new methodology for jointly modelling, learning, and controlling the dynamics of arterial traffic flows for accurate phase offset calculation. The approach introduces a nonlinear coupled oscillators model of the traffic network signalling system, along with a nonlinear control mechanism that allows it to capture complex flow patterns and unpredictable variations. This ensures a robust control of the oscillator-based model towards self-organization under dynamical demand changes based on the measurement of local traffic data. In order to benefit from realistic run-time performance under real traffic flows, the system is efficiently implemented in spiking neural networks, hence avoiding optimization routines and allowing it to operate in real time. Our real-world data experiments demonstrate that the methodology excels in minimizing typical traffic key performance indices when phase offset is calculated depending on the real-time demand measurements.

We conduct research and development of tools, models and infrastructure needed to interpret large amounts of clinical data and enhance cancer treatments and our understanding of the disease. To this end, this research direction serves as a bridge between the data, the engineer, and the clinician in oncological practice. Thus, knowledge-based predictive mathematical modelling is used to fill gaps in sparse data; assist and train machine learning algorithms; provide measurable interpretations of complex and heterogeneous clinical data sets, and make patient-tailored predictions of cancer progression and response.

Virtual Reality (VR) sensorimotor rehabilitation is still in its infancy but will soon require avatars, digital alter-egos of patients' physical selves. Such embodied interfaces could stimulate patients' perception in a rich and highly customized environment, where sensorimotor deficits, such as in Chemotherapy-Induced Peripheral Neuropathy, could be corrected. In such scenarios, motion prediction is a key ingredient for realistic immersion. Yet, such a task lives under hard processing latency constraints and the inherent variability of human motion. We work on neural network meta-learning systems exploiting the underlying correlations in body kinematics with the potential to provide, within latency guarantees, personalized VR rehabilitation. The unsupervised meta-learners are able to extract underlying statistics of the motion data by exploiting data regularities in order to describe the underlying manifold, or structure, of motion under sensorimotor deficits. We demonstrate, through preliminary experiments the potential of such a learning system for adaptive kinematics estimation in personalized rehabilitation VR avatars.

This research direction proposes the development of neural networks controller VR system for sport psychological (cognitive) and biomechanical training. By exploiting neuroscientific knowledge in sensorimotor processing, neural network-based learning algorithms and VR avatar reconstruction, our lab along with the other consortium partners target the development of an adaptive, affordable, and flexible novel solution for goalkeeper training in VR.

This research direction addresses the problem of monocular depth extraction in a comparative setup between traditional (frame-based) cameras and novel event-based cameras. The core goal of the project is the design and implementation of a test rig, a software framework for benchmarking algorithms, together with their calibration and validation. This first project sets the ground for upcoming research aiming at evaluating which is the best configuration of camera type and algorithm for monocular depth extraction for assistive devices.