Robotics & Control
I am interested in designing robots to tackle some of the modern challenges in manufacturing, infrastructure and healthcare. I am also interested in designing new nonlinear control techniques that balance robustness with optimality.
Biomechanics &
Healthcare devices
Understanding how animals and humans control their bodies fascinates me, and it is a major source of motivation for designing new rehabilitation and assistive devices.
Advanced Manufacturing
I am interested in both designing new manufacturing approaches and in improving existing one by using advanced sensing and control.
Sensing technologies
Understanding how processes work or designing efficient control/monitoring systems require advanced sensing. I am interested in both designing new sensors (especially at micro- and nano-scale) and in developing new data analysis algorithms.
I am interested in how robotics, automation and control techniques can be used to help humans in tasks that are either too dangerous, too time-consuming or requires large teams or high technical expertise of the operator.
For example, we have shown how pneumatically-actuated graspers can be designed to implement impedance control without requiring any sensor or real-time feedback. By carefully designing the geometry of such actuators we can tune their mechanical response, their interaction with objects and, ultimately, ensure safety. We are currently exploring how such the design and control of such actuators can be automatically optimised to use them for the next generation of assistive devices and lab manipulators.
We have also designed the first completely autonomous platform to perform 3D printing of construction materials on large (virtually unlimited) scale, by combining material deposition with autonomous ground robots. This will solve one of the main roadblocks of 3D printing of large structures, as it remove the necessity of having a frame to support the "printer" itself.
We are also interested in how robotics (and, more in general, mechatronics) can be used to automate a large variety of tasks, with projects ranging from remote monitoring and automation of hydroponic vertical farms, unconventional manipulators for chemistry labs, vision-based navigation and obstacle avoidance, environmental monitoring with robotic swarming, etc.
Finally, we are also interested in pushing the boundaries of the class of systems that can be controlled via rigorous optimal/robust control techniques. In fact, very often the systems we work with (in robotics, but also bio-mechanics, aerospace, hybrid testing etc...) present uncertainty and nonlinear behaviours that make them fall outside the hypotheses under which most of control theory has been developed. Therefore, we are interested in developing novel rigorous controller design strategies to enlarge the class of dynamical systems that can be controlled with rigorous/verifiable performance. We are particularly interested in how uncertainty can affect optimality and viceversa, We have shown that when an estimate of the uncertainty associated to a nonlinear dynamical model is known, an optimal controller can be designed while simultaneously guaranteeing stability in presence of such uncertainty. A typical example where such problem arise is in the aerospace industry, more specifically in flutter suppression problems, where there are significant (but often quantifiable) uncertainties related to airflow and structural nonlinearities in the wing.
Biological systems represent a very intriguing field where ideas from mechanics, nonlinear dynamics and control theory have to be mixed to correctly model the observed behaviours. Understanding how humans and animals move and interact with the surrounding environment is also a major inspiration for designing novel robots as well. My group mainly focus on problems related to animal locomotion and grasping, where we use optimal control ideas to understand the basic strategies that animals and humans can potentially adopt to coordinate the motion of their limbs or other body parts.
Past and current projects in this area include:
The availability of real-time data collected during additive manufacturing builds can be used to automatically detect faults (or, more in general, deviations) in the build process. The challenge here lies in the large quantity of data to be handled and processed (several hundreds of gigabytes per build), which requires development of algorithms that are fast and memory-efficient. By using statistical analyisis and machine learning, my group has shown defects such as powder shortage, under/over-melting, porosity can be reliably detected using only photodiode data collected during the build process. We are now extending the range of defects to be detected and the techniques to cover other additive manufacturing techniques. The ultimate goal here is to remove the necessity of post-build expensive tests and to implement in-build real-time control to correct defects during the build process itself.
We are also looking how robotics can help current manufacturing approaches, and even inspire the next-generation of manufacturing techniques. For example, we are developing soft and dexterous manipulators to be used in chemistry labs to accelerate the discovery of functional materials. We are also building small robotic devices that can self-assemble into functional products, with the final aim of creating "smart programmable atoms" that can algorithmically reconfigure in 3D objects and dynamically change the physical properties (stiffness, colour, shape, etc) of such objects when needed.
Modern technologies are largely based on the availability of data that allow us to monitor and control devices and physical processes. Sensors then play a critical role in providing such data-rich technological environment. Thanks to the availability of (relatively) cheap sensors, we now live in a data-rich world, which provides completely different control challenges compared to the traditional data-scarce scenario.
Together with several collaborators, we are developing several sensors to characterise material properties at the micro- and nano-scale, ranging from physics (mass-sensing and rheometry) to biology (biological tissue characterisation) to material science (surface topography). For example, we have introduced a novel dynamic imaging technique, called "auto-tapping" or "Dynamic Threshold Mode" (DTM) that exploits a nonlinear feedback loop to induce self excitation of the AFM probe. We have shown, both theoretically and experimentally, that such excitation mode makes it possible to achieve faster scan rate, still retaining the standard control setup that is present in most of the commercial AFMs and without increasing the tip-sample interation force. We have also shown that a similar setup can be successfully used to develop novel micro-rheometers and micro-mass sensing with a reliability and accuracy much greater than traditional sensors. The use of a self-excited cantilever also opens the possibility of using the same sensor in three different modalities by tuning only one parameter: analogue sensor, digital sensor ("threshold detector" and stable oscillator (insensitive to environmental conditions).
My group is also developing new ways of designing robust controllers for plants where physics-based models are not available or not accurate. By using sensor data, surrogate modelling, convex optimization and machine learning, robust controllers (in H2 and H-infinity sense, for example) can be designed for a wide variety of applications. For example, using such methodology we have successfully designed, implemented and validated several controllers for the automotive sector.
