Available Project Topics

Diploma Theses, Master Theses, Bachelor Theses, Research Projects and Student Reports

Projects Topics

The following project topics can be adapted for Diploma Theses, Master Theses, Bachelor Theses, Research Projects (Studienarbeiten), and Student Reports (Fachstudien). Contact Prof. Dr. Marco Aiello in case you have further questions or would like to submit your own proposals.

If you are interested in any of the following project topics, please send your inquiry to the following email: scsp@iaas.uni-stuttgart.de

Learning Deterministic Algorithms

An algorithm is a set of steps to follow in order to solve a given problem. Some algorithms are correct by design (sound and complete), while others are statistical in nature. E.g. a machine learning algorithm will use large amounts of data in order to classify with a given accuracy new data. The task of this research is to use machine learning to learn deterministic algorithms and (1) to evaluate the performance of the learning process, and (2) to compare the quality of the deterministic vs. the learned algorithm. Take for instance the Dijkstra algorithm for minimal paths on graphs. How can that algorithm be learned by examples of minimal paths given on a map? How many examples are necessary,  what will be the computational performance of the learned algorithm and the quality of its solutions?

Supervisors: Prof. Dr. Marco Aiello / Dr. Ilche Georgievski


Quantum Computing for Smart Energy Optimizations

The advent of practical Quantum Computing promises to speed up many computation processes, especially in the field of optimization. This means that some large scale simulations and optimizations that were unfeasible with traditional von Neumann architectures become practically available. The proposed research topic focuses on optimisations in the fields of energy distribution under modern models of Smart Grids [2]. The project is open ended. Possible goals include: feasibility of using quantum computing for energy optimization [1], encoding MILP energy optimisations as quantum problems, energy use cases implemented on quantum computers. The interested student can also propose his/her own topic within the sub-field of Quantum Computing for Smart Energy Optimizations.

Keywords: Quantum Computing, Smart Energy Systems, Energy Distribution, MILP


[1] Ajagekar, Akshay, and Fengqi You. “Quantum Computing for Energy Systems Optimization: Challenges and Opportunities.” Energy 179 (2019): 76–89. Crossref. Web.

[2] M. Aiello and G.A. Pagani (2016) How energy distribution will change: an ICT perspective, In Smart Grids from a Global Perspective, J. de Wilde, A. Beaulieu, and J. Scherpen (eds), Springer.

Supervisors: Prof. Dr. Marco Aiello / Daniel Vietz, M. Sc.


The Integration of Electric Vehicles in the Smart Grid

Electric Vehicles (eV) have a battery size that can vary from few KWh to almost 100 KWh. These are frequently connected to the power grid to charge. Several researchers have proposed the idea that these batteries can also be used to store grid electricity and provided it back to grid during peak demand periods. The research proposed here is to assess technical solutions to make the eV an integral part of the power grid, to study individual strategies based on dynamic pricing for incentivising eV integration, and to assess the economic viability of such approaches.


A. P. Lopes, F. J. Soares and P. M. R. Almeida, "Integration of Electric Vehicles in the Electric Power System," in Proceedings of the IEEE, vol. 99, no. 1, pp. 168-183, Jan. 2011.

Francis Mwasilu, Jackson John Justo, Eun-Kyung Kim, Ton Duc Do, Jin-Woo Jung, “Electric vehicles and smart grid interaction: A review on vehicle to grid and renewable energy sources integration,” in Renewable and Sustainable Energy Reviews, vol. 34, no. 6, pp 501-516, Jun. 2014.

Supervisor: Prof. Dr. Marco Aiello


Service Composition for Personal Assistants

Personal Assistants like Siri, Viv, Alex, Cortana, rely on online services to satisfy the requests of a user. Most often a request cannot be satisfied with just one service, but it needs the composition of multiple services. In this project you will investigate techniques to intelligently compose services  and you will implement prototypes connecting to Web services to satisfy complex user requests. 


Kaldeli, A. Lazovik, and M. Aiello (2016) Domain-Independent Planning for Services in Uncertain and Dynamic Environments, Artificial Intelligence, Elsevier, 236(7): 30-64.

Supervisor: Prof. Dr. Marco Aiello


Modelling and Simulating Research Funding Distribution in Computer Science

Today researchers heavily rely on research grants and subsidies to finance their research. The attribution of such subsidies is usually based on a peer-evaluation process that awards grants to the most promising proposals. In this project you will perform a statistical analysis of grant acquisition. You will provide a model to predict grant distribution and simulate various scenarios to identify the effects of specific policies on the probabilistic distribution of grants and on their impact on research and innovation. Bibliometric data for complex network analysis will be part of the developed model.

Supervisor: Prof. Dr. Marco Aiello


On the energy accountability of running software

Software running on data centres and personal computers is responsible for a large amount of energy consumption and consequent emissions worldwide. At the same time, it is very difficult to perform an accurate measure of energy consumption for individual software program and its individual components, this is also due to the current software engineering practices of layering and virtualisation. For this research project you are asked to look into energy models for software, energy accountability for software, comparing the running of the same algorithm on different platforms for measuring the energy consumption, etc.

Supervisor: Prof. Dr. Marco Aiello


Cellular Automata for Modelling Resource Allocation

Cellular automata are simple models that can represent complex even chaotic systems. The goal of this project is to provide simple models for resource allocation that help identifying efficient allocation rules. The project entails both a modelling and simulation component to it.

Supervisor: Prof. Dr. Marco Aiello


A Complex Network Analysis of Models to be Checked

Model Checking is the field concerned with the automatic process of checking whether a given model (or family of models) meets certain specifications. Model checkers work in various ways usually exploring large models explicitly represented as states or symbolically represented. Your task is to evaluate the impact of using complex network analysis techniques to improve and extend model checking approaches.

Supervisor: Prof. Dr. Marco Aiello


Quantum Computing for HTN Planning

Artificial Intelligence (AI) planning provides powerful techniques for searching and selecting actions that, when executed, achieve a given goal [1]. Finding solutions to planning problems is challenging: planning problems are often NP-hard or harder. Planning problems are currently addressed by either finding valid solutions using classical algorithms and selecting the appropriate ones (e.g., optimal) by hand or finding "good" solutions using heuristics. Even though these two approaches do not guarantee the generation of optimal solutions, both require substantial processing power.

The emergence of quantum computing promises an exploration of hard computational problems that are considered to be too demanding for classical computers [2, 3]. The hypothesis is that existing AI planning techniques could be enhanced by quantum algorithms that improve such techniques in the efficiency of finding suitable (e.g., optimal) plans, in the finding better plans (e.g., satisfy more constraints), and/or in a greater diversity of the plans found [4]. The possible goals of this project include exploring the feasibility of enhancing existing approaches on Hierarchical Task Network (HTN) planning [5] with quantum algorithms, encoding HTN planning problems as quantum problems, and demonstrating the benefits of quantum HTN planning on use cases. Interested students can also propose their own topics within the scope of the project.

Keywords: HTN planning, quantum computing


[1] M. Ghallab, D. S. Nau, and P. Traverso, Automated planning: Theory & practice. Morgan Kaufmann Publishers Inc., 2004.

[2] VV. N. Smelyanskiy, E. G. Rieffel, S. I. Knysh, C. P. Williams, M. W. Johnson, M. C. Thom, W. G. Macready, and K. L. Pudenz, "A near-term quantum computing approach for hard computational problems in space exploration", arXiv: Quantum Physics, 2012.

[3] R. Desimone, P. Warburton, Y.-L. Fang, A. Montanaro, S. Piddock, S. Yarkoni, A. Mason, C. White, and T. Popa, "Quantum computing algorithms for optimised planning and scheduling applications (QCAPS)", Plantagenet Systems Ltd, QCAPS/133087/Final Report/D008/v1.0, 2018.

[4] R. Biswas, Z. Jiang, K. Kechezhi, S. Knysh, S. Mandra, B. O'Gorman, A. Perdomo-Ortiz, A. Petukhov, J. Realpe-Goomez, E. Rieffel, D. Venturelli, F. Vasko, and Z. Wang, "A NASA perspective on quantum computing: Opportunities and challenges", Parallel Comput., 64, 81-98, 2017.

[5] I. Georgievski and M. Aiello, "HTN planning: Overview, comparison, and beyond", Artif. Intell., vol. 222, pp. 124-156, 2015.

Supervisor: Dr. Ilche Georgievski


Demand-Side Management via AI Planning

Commercial buildings are characterised by high operational costs and high CO2 emission levels due to the high demand for energy [1]. The management of energy demand, also known as demand-side management, allows buildings to make informed decisions about their energy consumption, but it also helps energy providers to shape load demand profiles [2]. However, current building management systems are capable of controlling heating, ventilation and air conditioning, lighting and other aspects using specific strategies limited by basic scheduling functions [3]. On the other hand, with the proliferation of the Internet of Things (IoT) deployed in building, building management systems can be transformed into systems employing intelligent demand-side management. In other words, buildings equipped with various sensors and actuators have the potential to become intelligent and provide occupants with a more comfortable experience, reduced operational costs, and improved energy efficiency.

This calls for developing approaches that can plan the energy demand of buildings intelligently. In this context, Artificial Intelligence (AI) planning offers powerful techniques for automated and dynamic selection and organisation of actions that, when executed, achieve a given objective [4]. AI planning could be helpful for buildings to intelligently plan their demand by computing effective plans or schedules of device actions.

The aim of this project is to propose and develop an approach for demand-side management via AI planning. This involves defining the problem of demand-side management as a domain-independent planning problem, modelling the planning problem using a standard syntax, such as PDDL [5], solving the planning problem using existing AI planners, and evaluating the efficiency and effectiveness of the proposed approach.

Keywords: AI planning, demand-side management, Internet of Things, office buildings


[1] D. D'Agostino, B. Cuniberti, and P. Bertoldi, Energy consumption and efficiency technology measures in European non-residential buildings," Energy and Buildings, vol. 153, pp. 2-86,2017.

[2] P. Palensky and D. Dietrich, Demand side management: Demand response, intelligent energy systems, and smart loads," IEEE Transactions on Industrial Informatics, vol. 7, no. 3, pp. 381-388, 2011.

[3] I. Georgievski, T. A. Nguyen, F. Nizamic, B. Setz, A. Lazovik, and M. Aiello, Planning meets activity recognition: Service coordination for intelligent buildings," Pervasive and Mobile Computing, vol. 38, no. 1, pp. 110-139, 2017.

[4] M. Ghallab, D. S. Nau, and P. Traverso, Automated planning: Theory & practice. Morgan Kaufmann Publishers Inc., 2004.

[5] M. Fox and D. Long, PDDL2.1: An extension to PDDL for expressing temporal planning domains," Journal of Artificial Intelligence Research, vol. 20, no. 1, pp. 61-124, 2003.

Supervisor: Dr. Ilche Georgievski


Ubiquitous Computing Domain: Benchmark for Planning

Ubiquitous computing aims at creating ambience in which one's experiences and quality of life are improved by monitoring and assisting people using Internet of Things and intelligence in coherence. Artificial Intelligence (AI) planning provides a sophisticated form of intelligence [1] and appears to be highly suitable for ubiquitous computing [2]. Even though AI planning has been used to solve problems in many planning domains in ubiquitous computing, it is still unknown how difficult such problems are and the amount of resources those problems require.

Knowing the complexity of planning in specific domains gives an opportunity to outline the speed and length of plans generated by AI planners. Since it is impossible to analyse the complexity of every planning domain in ubiquitous computing, we can look at the domains from a broader perspective - a single and general ubiquitous computing planning domain (benchmark domain) [3]. Benchmarking should help answer how AI planners perform on such a benchmark domain, how their performance compare on other benchmark domains, what are the main reasons for differences in performance, and what could be learned from other domains to potentially improve the ubiquitous computing planning domain [4]?

The aim of this project is to define and model a benchmark planning domain for ubiquitous computing. This involves defining a general ubiquitous computing planning domain and three specialisations of it (e.g., non-temporal, temporal, offices, houses, etc.), deciding which tasks, actions, and objects should be part of the general domain, modelling the domain in PDDL, and possibly testing a selection of AI planners on the PDDL version of the benchmark domain.

Keywords: Ubiquitous computing, planning benchmarks, planning domains


[1] M. Ghallab, D. S. Nau, and P. Traverso, Automated planning: Theory & practice. Morgan Kaufmann Publishers Inc., 2004.

[2] I. Georgievski, "Coordinating services embedded everywhere via hierarchical planning", Ph.D. dissertation, University of Groningen, October 2015.

[3] I. Georgievski and M. Aiello, "Automated planning for ubiquitous computing", ACM Comput. Surv., vol. 49, no. 4, pp. 63:1-63:46, 2016.

[4] J. Hoffmann, S. Edelkamp, S. Thiebaux, R. Englert, F. dos Santos Liporace, and S. Trüg, "Engineering benchmarks for planning: The domains used in the deterministic part of ipc-4" J. Artif. Int. Res., vol. 26, no. 1, pp. 453-541, 2006.

Supervisor: Dr. Ilche Georgievski


HTN Planning for Automated Deployment and Dynamic Reconfiguration of Cloud Applications

Modern software applications are increasingly deployed and distributed on infrastructures in the Cloud. Before the deployment process happens, these applications are being manually - or with some predefined scripts - composed from various smaller interdependent components. With the increase in demand for, and complexity of applications, the composition process becomes an arduous task often associated with errors and a suboptimal use of computer resources. Dynamic reconfiguration, which is the process of reconfiguring an application while it is running, only aggravates the problem even further [1]. Such processes can be alleviated by automatically and dynamically composing and reconfiguring Cloud applications , especially when clear semantics exists that describe the deployment and reconfiguration problems, such as the Aeolus formal model [2].

Artificial Intelligence (AI) planning provides techniques for searching and selecting actions that, when executed, achieve a given goal [3]. In the context of the Cloud, AI planning can provides powerful methods for searching in large and complex Cloud infrastructures to find "good" compositions and reconfigurations of Cloud applications. Additionally, AI planning can be used to handle the Cloud uncertainty (e.g., failures of hardware resources), find deployments optimal with respect to the use of computer resources, etc. Among AI planning techniques, Hierarchical Task Network (HTN) planning appears suitable for solving deployment problems [4].

The aim of this project is to study and solve both deployment and reconfiguration problems using HTN planning. The project should build upon [4]. The project tasks include:

  • Refine the definition of the Aelous-based deployment problem as an HTN planning problem (to improve performance);
  • Define the Aelous-based reconfiguration problem as an HTN planning problem;
  • Model such HTN planning problems using HPDL and/or SHOP2;
  • Evaluate the approach using one or more existing AI planners;
  • Compare the performance of existing AI planners among themselves and with Metis [5].

Keywords: Automated deployment, dynamic reconfiguration, cloud applications, HTN planning


[1] N. Arshad, D. Heimbigner, and A. Wolf, "Deployment and dynamic reconfiguration planning for distributed software systems," in International Conference on Tools with Artificial Intelligence, ser. ICTAI'03, 2003, pp. 39-46.

[2] R. Di Cosmo, J. Mauro, S. Zacchiroli, and G. Zavattaro, "Aeolus: A component model for the cloud," Information and Computation, vol. 239, pp. 100-121, 2014.

[3] M. Ghallab, D. S. Nau, and P. Traverso, Automated planning: Theory & practice. Morgan Kaufmann Publishers Inc., 2004.

[4] I. Georgievski, F. Nizamic, A. Lazovik, and M. Aiello, "Cloud ready applications composed via htn planning," in IEEE International Conference on Service Oriented Computing and Applications, ser. SOCA'17, 2017, pp. 23-33.

[5] T. A. Lascu, J. Mauro, and G. Zavattaro, "A planning tool supporting the deployment of cloud applications," in International Conference on Tools with Artificial Intelligence, ser. ICTAI'13, 2013, pp. 213-220.

Supervisor: Dr. Ilche Georgievski


Trust-Based HTN Planning for Automated Vehicles

Human trust is an important factor to consider when planning routes for automated vehicles. Human trust can be seen as a delegation of responsibility for actions to automation and willingness to accept risk, including possible harm. Especially relevant for automated vehicles is cognitive trust, which accounts for both the human experience and the subjective judgment about preferences and mental states [1]. The human's willingness to rely on automation or the decision to delegate is based on a subjective evaluation of the capability of automation to perform a particular task [2]. For example, if the human has lower trust in the capability of the automated vehicle for safely navigating urban streets with pedestrians constantly crossing as opposed to freeways, the human may prefer a free-way despite the longer distance [3].

Trust in automation changes over time during interaction with automation and can be influenced by many factors, such as the automation's reliability, predictability, and transparency [4,5]. For our project, we assume that trust evolves depending on the vehicle's capability of handling an incident. Intuitively, the trust may increase when the automated vehicle successfully handling an incident, and the trust may decrease if the automated vehicle fails to handle an incident. At least three types of typical incidents that may occur on a road should be considered: a pedestrian crossing the road, an obstacle ahead of the lane, and an oncoming truck in the neighbouring lane.

The aim of this project is to investigate how to model trust and incorporate it into planning routes for automated vehicles using Artificial Intelligence planning with Hierarchical Task Networks (HTNs) [6]. If there would be a reward for improved safety and user experience of automated vehicles, what should be it? The goal of the approach could be to compute plans with route choices that maximise the expectation of the cumulative reward. We are particularly interested in how utility-based HTN planning can be used for this [7].

Keywords: Automated vehicles, trust, HTN planning, utilities


[1] R. Falcone and C. Castelfranchi, Social Trust: A Cognitive Approach. Kluwer Academic Publishers, 2001, pp. 55--90.

[2] J. D. Lee and K. A. See, Trust in Automation: Designing for Appropriate Reliance, Human Factors, vol. 46, no. 1, pp. 50-80, 2004.

[3] S. Sheng, E. Pakdamanian, K. Han, Z. Wang, J. Lenneman, and L. Feng, Trust-Based Route Planning for Automated Vehicles, in ACM/IEEE International Conference on Cyber-Physical Systems, ser. ICCPS '21, 2021, pp. 1-10.

[4] P. A. Hancock, D. R. Billings, K. E. Schaefer, J. Y. C. Chen, E. J. de Visser, and R. Parasuraman, A meta-analysis of factors affecting trust in human-robot interaction, Human Factors, vol. 53, no. 5, pp. 517-527, 2011.

[5] K. E. Schaefer, J. Y. C. Chen, J. L. Szalma, and P. A. Hancock, A Meta-Analysis of Factors Influencing the Development of Trust in Automation: Implications for Understanding Autonomy in Future Systems, Human Factors, vol. 58, no. 3, pp. 377-400, 2016.

[6] I. Georgievski and M. Aiello, HTN planning: Overview, comparison, and beyond, Artif. Intell., vol. 222, pp. 124-156, 2015.

[7] E. Alnazer, HTN Planning with Utilities, Master's thesis, University of Stuttgart, 2019.

Supervisors: Dr. Ilche Georgievski / M.Sc. Ebaa Alnazer


AI Planning Under Uncertainty for Sustainable Buildings

With the increment of CO2 emission, the world energy consumption, and the diversity in energy sources (finite and renewable), the goals of building management systems are widened to include cutting the energy consumption; hence preserving finite resources, using non-carbon sources when possible to lower the CO2 footprint, and lowering costs for consumers and businesses while preserving users comfort [2]. This is especially true for buildings that are connected to a smart grid, which makes it possible to include renewable sources and provides dynamic pricing and energy offers coming from competing providers [4]. The advent of the Internet of Things (IoT) offers significant opportunities to improve the limited control capabilities offered by current building management systems [5]. In other words, equipping commercial buildings with different sensors and actuators allows management systems to make informed and "smart" decisions to control the environment and adjust it towards satisfying predefined objectives, such as energy saving. This "smartness" implies automating the decision-making process. Artificial Intelligence (AI) planning fits perfectly for this purpose since it offers automation techniques to intelligently search and compose a sequence of actions, i.e., plans when executed, can achieve a predefined set of goals [6].

Planning in such domains is challenging due to several types of uncertainty, e.g., weather forecasting, demand forecasting, and market prices forecasting [3, 2]. Consider the following example. A smart building is connected to a smart grid and has the goal of maintaining economical and environmental sustainability. The building can get its electricity from two different sources; either from stored electricity generated locally by renewable resources or from electric utilities with varying prices and energy offers. The choice of whether to consume the locally generated electricity or the purchasing it from outside sources should be made under uncertainty about the future market prices and the future weather forecast. In many approaches, AI planning problems are modelled with uncertainties about the action outcomes in mind; hence the resulting plans in such approaches are of a probabilistic nature [7, 1].

The aim of this project is to develop an AI planning-based approach in such domains. This involves defining the entities that are part of these domains, defining the types of uncertainty that can arise in these domains, modelling the AI planning problem by using a standard syntax that can model some sort of uncertainty, e.g., PPDDL [8], solving the planning problem using existing AI planners that can deal with uncertainties, and evaluating the efficiency and effectiveness of the proposed approach.

Keywords: sustainable buildings, Internet of Things, AI planning


[1] Blythe, J.: An overview of planning under uncertainty. Artificial Intelligence Today, pp. 85-110 (1999).

[2] Fiorini, L., Aiello, M.: Energy management for user’s thermal and power needs: A survey. Energy Reports 5, 1048-1076 (2019).

[3] Fiorini, L., Aiello, M.: Predictive multi-objective scheduling with dynamic prices and marginal CO2-emission intensities. In: Proceedings of the Eleventh ACM International Conference on Future Energy Systems. pp. 196-207 (2020).

[4] Georgievski, I.: Coordinating services embedded everywhere via hierarchical planning. Ph.D. thesis, University of Groningen (2015).

[5] Georgievski, I., Nguyen, T.A., Nizamic, F., Setz, B., Lazovik, A., Aiello, M.: Planning meets activity recognition: Service coordination for intelligent buildings. Pervasive and Mobile Computing 38, 110-139 (2017).

[6] Ghallab, M., Nau, D., Traverso, P.: Automated Planning: theory and practice. Elsevier (2004).

[7] Koenig, S., Simmons, R.G.: Risk-sensitive planning with probabilistic decision graphs. In: Principles of Knowledge Representation and Reasoning. pp. 363-373. Elsevier (1994).

[8] Younes, H.L., Littman, M.L.: PPDDL1. 0: An extension to PDDL for expressing planning domains with probabilistic effects. Techn. Rep. CMU-CS-04-162 2, 99 (2004).

Supervisor: M.Sc. Ebaa Alnazer


Cluster Scheduling in Data Centers

Cluster schedulers are a key component of managing the workload and the computational resources of data centers. Schedulers make scheduling decisions by gathering the information about the load and the queues on compute nodes as well as the available resources by communicating with the resource managers. Over the past years, a number of schedulers such as Borg [1], Mesos [2], Apache Yarn [3], and Slurm [4] have been developed. These schedulers are different in many aspects and regarding various key features like cluster type, language support, job migration, and resource allocation policies. The possible goals of this project include scientific evaluation of various resource allocation policies used in current schedulers and exploring the compatibility of different schedulers for GPU enhanced clusters. Interested students can also propose their topics within the scope of the project.

Keywords: Cluster scheduling, Data centers


[1] Verma, A., Pedrosa, L., Korupolu, M.: Large-scale cluster management at Google with Borg. In: Proceedings of the Tenth European Conference on Computer Systems. (2015).

[2] Hindman. B., Konwinski, A., et al.: Mesos: A platform for fine-grained resource sharing in the data center. In: NSDI. (2011).

[3] Vavilapalli. V., Murthy, A., et al.: Apache hadoop yarn: Yet another resource negotiator. In: Proceedings of the 4th annual Symposium on Cloud Computing. (2013).

[4] Yoo. AB., Jette, MA., et al.: Slurm: Simple linux utility for resource management. Workshop on job scheduling strategies for parallel processing. Springer, Berlin, Heidelberg. (2003)

Supervisor: Dr. Kawsar Haghshenas


This image shows Marco Aiello

Marco Aiello

Prof. Dr.

Head of Department

This image shows Elisabeth Ibach

Elisabeth Ibach



This image shows Ilche Georgievski

Ilche Georgievski


Lead of Smart Energy Systems Research Area

[Photo: Ilche Georgievski]

To the top of the page