Intentionally, we combine and apply expertise from multiple robotics specialities that normally act in isolation. This includes more traditional and well-established research in perception, communication and control, as well as more advanced and pioneering topics such as machine learning, deep learning and reinforcement learning for autonomous decision making. It also includes bio-inspired, self-organised and evolutionary systems with the potential to achieve life-long and adaptable autonomy.
Cross-fertilisation across these fields produces cutting-edge new technologies and intellectual property that will enable our autonomous systems to perform operations in non-trivial environments such as unstructured land, air, and water. Some of these operations include:
- Three-dimensional mapping and navigation of cluttered multi-floor environments to locate victims of disaster
- Area coverage with single- and multi-robot systems for surveillance and monitoring of large indoor or outdoor locations
- Autonomous driving off-road with minimal signalling and infrastructures and potentially non-perfect perception
- Reactive and deliberative autonomous control of large swarms of unmanned aerial vehicles that perform patrolling and scans of large territories
- Large-scale sensor networks with limited connectivity requirements
- Bio-inspired robotic solutions to restore damaged eco-systems on land, on water and under water.
Our research extends from fundamental and theoretical studies to prototype development and production of intellectual property. We use mathematical and conceptual models. We then move to simple or more sophisticated and detailed simulations. We use existing robotic platforms, integrated with novel sensing and actuation mechanisms, and develop novel hardware and software platforms for different needs.
We develop prototypes to benefit the worldwide robotics community, and we participate in international robotics competitions and challenges. Our institute has access to world-leading infrastructure and laboratories to support all these activities.
Our starting point is state-of-the-art sensor technology, computer vision, and tele(remote) -communication. We apply these technologies to boost robotics solutions.
On perception, we work on 2D and 3D computer vision techniques, working at the interface between the vision classical school and the latest deep-learning methodologies. This enables real-time perception and sensor fusion of landmarks, in either large or constrained environments, to build a map, or to re-build the 3D profile of a complex surface (e.g., the seabed or a shipwreck).
We also use vision and other sensor modalities to perform localisation of robots in scenarios where global positioning systems are not available and to perform relative localisation of other robots to enable swarm methodologies.
We begin with the assumption that human-built infrastructure, such as the Internet, are not always available. In such scenarios, we use 5G and beyond 5G advancements. Assuming only mobile ad-hoc networking, our robotics systems can build temporary communication networks from scratch, to be used in extreme circumstances such as disaster recovery.
We develop self-organised routing protocols to enable reliable or fast communication infrastructure. These communication mechanisms work in situations where the nodes of the network (the robots or other contingent devices) are not static and potentially may face constantly changing neighbours, in a controlled or uncontrolled fashion.
Building on existing research into control and machine learning, we extend control and decision making into challenging new domains.
We develop novel control algorithms for single and multi-robot systems (swarms) that are either flying or swimming under the sea. Our scenarios do not assume full nor perfect knowledge of the environment, as our robots cannot operate with precise positioning and therefore perfect state information.
As such, we extend traditional control approaches to cope with noisy and unknown environments. We work with robots that are custom designed for specific applications, or are the result of an evolutionary process and not designed by humans. These constraints impose challenges to existing control approaches and therefore motivate further research advancements.
We use exciting and novel control theory proofs (e.g., proving stability or robustness) to verify new robotics platforms used to obtain certifications for safety of operation.
Research in decision making typically concerns the problem of increasing the autonomy of robotics systems in terms of taking the appropriate actions in a previously unseen and potentially dynamic environment.
We use deep reinforcement learning and other machine learning methods to design advanced single- and multi-robot systems for unseen environment. For example, we are developing a decision-making agent able to drive a car or jeep autonomously in an unforeseen and unstructured environment such as the desert.
In the context of multi-robot systems, we use collaborative learning methods hybridized with other robotics methods to perform area coverage of an unseen environment, in which the geometry is not known, sensing is limited and only local, and different areas require different degrees of coverage that could change over time.
Our long-term goals require methodologies that go beyond classic robotics and machine learning. This requires unconventional research that examines how nature solves a number of real-world problems through biological systems that have evolved over millions of years.
The biology of animal organs, from sensorial such as eyes and ears, to manipulative and locomotional such as octopus tentacles and gecko toes, can be used as inspiration to design novel solutions for robots using soft or organic materials and joints.
An optimisation process can be reproduced in silicon in order to simultaneously co-design new robots and robot behaviours from first principles and basic building blocks that resemble cells. As well, we learn from principles of self-organisation that drive collective behaviours such as complex molecules generated from simpler ones, laser beam synchronisation, firefly synchronisation, bird flocking, locust swarming, and termites building huge and complex architectures due to the microscopic actions of a multitude of individuals.
This research studies the principles and mechanisms that make biological organisms, evolution, and self-organisation so effective. We want to build strong foundations around these principles to build a novel engineering field able to reproduce only the positive outcomes of natural processes. This involves driving these processes in a way that avoids nature’s long trial and error process.
Within this framework, we work in a number of ambitious engineering processes. We use evolutionary methods to simultaneously design the body and the brain of robots that have to solve tasks underwater or in the air. In the context of self-organisation, we design simple mechanisms for controlling very large groups of robots to minimise the hardware requirements for each robot, thus enabling mass production with minimised cost. Other mechanisms seek to coordinate motion on land and air as building blocks for more complex collective behaviour.