TU Delft’s five ERC Consolidator Grant projects are:
Shadi Sharif Azadeh, Civil Engineering & Geosiences (CEG)
TRANSFORM - Transport resilience and adaptive networks: a holistic framework for multi-scale optimization under uncertainty in a rapidly changing mobility environmentMultimodal passenger mobility networks in urban areas are subject to constant change due to technological innovations, new business models and expansion of the current infrastructure to accommodate growing number of travellers and adapt to innovative mobility solutions.
Consequently, there are several major concerns to be addressed while adapting the current transport network to an evolving environment:
i) given limited space and constantly growing demand in urban areas transport operators and public authorities need to guarantee that long-term costly investments will improve mobility service level for passengers, and result in efficient usage of resources for service providers;
ii) multimodal transport management is highly affected by both local and global disruptions (uncertainties) such as accidents, infrastructural malfunctions, and extreme weather and climate conditions. Shortand long-term asset management require identifying these uncertainties and accurately quantifying them to maximize efficient usage of available resources, and finally, iii) travellersbehavioural response to sudden disruptions and their long-term adaptation to new mobility solutions must be measured to guarantee the usability of mobility services in a changing environment.
TRANSFORM will develop a breakthrough smart -estimate-then-optimizeframework for a robust multi timescale asset management of multimodal transport systems in an uncertain environment. TRANSFORM will introduce cutting-edge predictive models coupled with novel adaptive real-time optimization methods that leverage real-time data analytics and advanced optimization algorithms, setting this approach apart from traditional asset management frameworks. TRANSFORM will assess travellersbehavioural adaptation to new mobility solutions, enabling the implementation of targeted demand steering strategies to ensure high system usability and passenger satisfaction.
TRANSFORM entails multidisciplinary methodologies to significantly contribute to sustainable urban mobility systems :
i) conceptualizing multimodal transport with three players: multimodal mobility service supplier, infrastructure operators and consumers,
ii) dynamically quantifying uncertainties,
iii) creating behaviorally informed demand adaptation strategies to model supply and demand interplay in real-time,
iv) iterative optimization under uncertainty, and
v) developing a multi timescale multi-agent simulation.
Optimization models under uncertainties, multi timescale uncertainty quantification, and adaptive behavioral framework all benefit from econometric analysis that necessitates operations research and decision theory expertise. Only when all these come together, I can show the impact of my innovations.
TRANSFORM has been designed based on the transportation domain knowledge linked to social science and humanities. The nature of the transportation systems brings the scientific challenge.
View the personal page of Shadi.
Louise Nuijens, Civil Engineering & Geosiences (CEG)
QUASI - Stormy Atmospheres over Quiescent Waters: Dynamical Implications of Fine-scale Air-Sea Interaction The atmosphere and ocean exhibit natural fluctuations at many scales. Compellingly, their responses to each other-s cues, felt through fluxes at the interface, may be taking place at much finer scales. At the air-sea interaction (ASI) submesoscale, ranging from about 200 m to 200 km, both air and water are filled with beautiful heterogeneous structure. Unraveling how such fine-scale structure drives air-sea exchange, leading to different dynamics in both fluids, is an emerging challenge for climate science.My ambition is to expose dynamical impacts and primary mechanisms of submesoscale interaction between atmospheric moist convection and tropical quiescent waters. With QUASI, I will use Earth-s stormiest and largest tropical lake as an ocean mixed-layer analog, avoiding the complexity of a deep wavy ocean. Lake Victoria offers the ideal field lab to study sensitive waters and strong atmospheric signals in the form of near-daily convective storms triggered by warm waters and mesoscale circulations.
QUASI will realize a multi-sensor multi-buoy network on the lake to measure spatial, cross-interface, high-resolution observations of the near-surface atmosphere and water for over a year. The observations force ocean mixed-layer models to uncover important variability and inspire and inform advanced atmospheric large-eddy and storm-resolving simulations in uncoupled and coupled configurations.
QUASI will identify key atmospheric drivers and scales of heterogeneity in air-sea fluxes, expose responses of surface water temperature and mixing, infer dynamical implications of treating water as a static homogeneous surface, and study the impact of submesoscale ASI on regional weather patterns and extremes. While inherently challenging, QUASI delivers rare cross-disciplinary evidence critical for advancing a new era of coupled high-resolution models needed to unravel surprising energy pathways in our climate system.
Dante Muratore, Electrical Engineering, Mathematics and Computer Science (EEMCS)
LOOK - Attention-Based Compressive Readout Chips for Massively Parallel Brain-Computer Interfaces Brain-computer interfaces (BCIs) can revolutionize society by restoring lost function and augmenting human capabilities. Next-generation BCIs must record large neuron populations at single-cell and cell-type resolution to speak the natural language of the brain. This requires substantial advances in large-scale low-noise readout electronics while addressing power, area, and data transmission bottlenecks.LOOK will develop radically new low-power and low-area -nearly losslesscompressive readout chips for massively parallel single-cell BCIs. I will tackle a fundamental challenge when compressing noisy signals: lossless compression provides insufficient data reduction; lossy compression achieves higher data reduction but misses low amplitude events. Inspired by the retina, LOOK will combine lossy and lossless compressive readouts with -attentionmechanisms to achieve high compression ratio and accuracy at the same time.
Unlike conventional compression that exploits inherent redundancy and structure within the data, LOOK will use prior knowledge of neural signaling to dynamically determine which parts of the signal are most important and, therefore, selectively apply lossless compression to only those subsets while aggressively compressing the rest.
To develop this technology, I will adopt a circuit-algorithm co-design approach, where the development of compression and attention algorithms will be tightly integrated with the design of low-power recording chips. This approach will be validated with human pre-recorded neural data and in vitro neural recording experiments to show its applicability to current speech BCIs and future natural resolution interfaces.
LOOK will surpass traditional accuracy/power trade-offs in compressive readouts to enable massively parallel, single-cell resolution BCIs. The ground-breaking nature of this project will also impact other applications where power-constrained autonomous multi-sensor arrays record noisy, sparse signals.
View the personal page of Dante.
Sergio Grammatico, Mechanical Engineering (ME)
ARGON - Data Driven Game Theoretic Control for Constrained Systems With this prestigious grant of 2 million euros, Sergio aims to develop novel control systems that reduce reliance on model-based knowledge and instead leverage empirical data.Current control systems, such as those used in traffic management, robotics, and power grids, are mostly model-based. They operate with a -digital twinof reality, making decisions based on predefined models rather than empirical and real-time evidence. In practice, modern engineering infrastructures consist of interconnected subsystems, often self-interested and sharing limited information, which makes accurate modelling significantly harder. -When these models are inaccurate, consequences can be severe,- Grammatico explains, -from traffic congestion to large-scale failures such as power blackouts, where model unawareness and competition between subsystems play an important role.-
Quicker adaptation
Grammatico-s project, called ARGON, will focus on developing control systems that rely as little as possible on models and instead use empirical data and real-time measurements. This approach allows controllers to adapt faster to changing conditions and reduces the risks associated with incorrect model assumptions.
As an example of the practical implications of his project, Grammatico mentions human-robot collaboration. -In this case, multiple agents, industrial robots and humans, work together in a shared environment. Traditionally, robots are designed based on models of human behaviour, which may not reflect actual preferences or needs for human comfort. By using operational data instead, industrial robots can adapt quickly and effectively to human operators.-
Proactive and reactive control
Because it is impossible to know all parameters of human behaviour and in general of our increasingly complex systems, Grammatico believes that focusing on empirical data is essential. His project aims to develop both the theoretical foundations of model-free control design in competitive environments and algorithms that can efficiently solve decision and control problems in competitive environments. The project will develop two main lines of research: proactive (feedforward) control, which anticipates future system behaviour, and reactive (feedback) control, which responds in real time to observed changes.
-I believe that in the future, the design of autonomous control systems will combine model-based and data-driven approaches,- Grammatico concludes. -The best trade-off is to be found by exploiting what we know from the physics while listening to the empirical data-.
View the personal page of Sergio.
Arjen Jakobi, Applied Sciences (AS)
INFLAMMAZOOM - Illuminating the structural mechanism of non-canonical inflammasomes Inflammasomes are important components of the innate immune system. These multi-protein complexes help limit the growth of pathogens inside cells and coordinate inflammatory signals. A special class of inflammasomes detects bacterial lipopolysaccharides (LPS), large molecules consisting of a lipid and a sugar component that form a key part of the outer membrane of Gram-negative bacteria. One example are the Salmonella bacteria that can cause typhoid fever. This recognition triggers a chain reaction that ultimately causes the infected cell to burst open, allowing the bacteria to be cleared.Although these protein complexes play a crucial role in immune defence, we still know little about how they work. This is because they bind only transiently and in complex assembly patterns to the highly complex outer membrane of Gram-negative bacteria. As a result, they are extremely difficult to study with existing structural biology methods, and the details of their assembly, activation and specificity have remained largely unclear.
In this project, Arjen Jakobi will investigate the structure and function of these inflammasomes using new electron cryo-microscopy methods that he will specifically develop to study dynamic protein complexes on membranes. To follow these processes, Jakobi will design techniques capable of capturing rapid changes in protein structures. These methods must reveal structural changes on the millisecond timescale and do so at atomic resolution. This will allow him and his team to generate a series of static snapshots showing, step by step, how an inflammasome forms and becomes active. Ultimately, these techniques should also work inside cells, enabling fast molecular processes to be visualised in their natural environment.
Understanding these fundamental mechanisms is essential. Excessive inflammasome activity can contribute to serious conditions such as sepsis and damage to the blood-brain barrier. The new methods developed in this project will also be broadly useful for studying other large protein complexes that function on complex biological membrane surfaces.
View the personal page of Arjen.