Abolfazl Lavaei

Abolfazl Lavaei 

Abolfazl Lavaei
Ph.D. Candidate

Department of Electrical and Computer Engineering
Technical University of Munich
3018 in building 2906
Karlstr. 45/47
80333 München

Tel: +
E-Mail: lavaei@tum.de


Abolfazl Lavaei is currently a PhD candidate in the Hybrid Control Systems Lab, Department of Electrical and Computer Engineering at the Technical University of Munich (TUM), Germany, since November 2016. He is also a Munich Aerospace Doctoral Scholarship holder under the research group of Autonomous Flight. Prior to joining TUM, he received the M.Sc. degree in Aerospace Engineering with a major in Flight Dynamics and Control from University of Tehran, Iran, in 2014. For his Master's work, he has received the Best Graduate Student Award in all fields of study in Faculty of New Sciences and Technologies at University of Tehran with the GPA of 20/20. He was also the recipient of several prestigious PhD scholarships from different top-ranked universities. He was also selected as one of the top three finalists for the IFAC Young Author Award at the 15th IFAC Symposium on Large-Scale Complex Systems: Theory and Applications (LSS), 2019. During his PhD studies, he has been working on "Automated Verification and Control of Large-Scale Stochastic Cyber-Physical Systems via Compositional (Formal) Methods". More specifically, his current research focuses on developing abstraction-based synthesis techniques for stochastic cyber-physical systems (e.g., traffic networks) in an automated as well as formal fashion in order to enable agents to make their own decisions without direct human involvement (e.g., autonomous driving). He is a scholar of Munich Aerospace Research Group as well as DLR Graduate Program.

Research Area

Autonomy is certainly one of the main themes of the 21st-century technology. In the near future, we expect to see fully autonomous vehicles, aircrafts, and robots, all of which should be able to make their own decisions without direct human involvement. Although this technology theme provides many potential advantages, e.g., fewer traffic collisions, reduced traffic congestion, increased roadway capacity, relief of vehicle occupants from driving, and so forth, guaranteeing safety and reliability of such networks in a formal as well as time/cost-effective way is the main challenging objective in the study of those systems.

In this respect, one promising approach is to first drive infinite abstractions (model order reductions) of such complex networks, and then construct finite abstractions (a.k.a. finite Markov decision processes (MDPs)) from the reduced-order versions of the original models. By leveraging those constructed finite MDPs as substitutions of the original ones in the controller design process and developing computational tools for stochastic systems, one can synthesize controllers in an automated as well as formal fashion to satisfy some high-level specifications, e.g., those expressed as linear temporal logic (LTL) formulae, which are difficult (if not impossible) to enforce with classical control theory approaches. Accordingly, one can refine the controllers back (via an interface map) to the original models with providing guaranteed error bounds on probabilistic output trajectories. Unfortunately, almost all the existing techniques on the construction of finite abstractions in a monolithic manner suffer severely from the curse of dimensionality: the complexity exponentially grows as the number of state variables increases. To mitigate this issue, one promising solution is to consider the large-scale stochastic system as an interconnected system composed of several smaller subsystems, and provide a compositional framework for the construction of (in)finite abstractions for the given system using the abstractions of smaller subsystems. Hence, the main goal of my current research is to develop compositional formal methods for automated (push-button) verification and control of large-scale stochastic cyber-physical systems which are absolutely crucial in many safety-critical applications such as autonomous driving.

Bachelor & Master Thesis Opening

Please contact me if you are interested in doing your Bachelor or Master thesis under my supervision. The current openings revolve around:

  • Implementation of controller synthesis in C++/Python with application to "autonomous driving";

  • Formal controller synthesis of unknown stochastic hybrid systems via "model-free reinforcement learning";

  • Temporal logic verification and synthesis of unknown stochastic hybrid systems via "data-driven optimization".

If you have also your own idea for a thesis topic, feel free to contact me.

Former Supervision

  • Winter Semester 2018: Master Thesis. Topic: A Tool in C++ on Abstraction-based Synthesis of Stochastic Systems.

  • Summer Semester 2018: Advanced Seminar on Cyber-Physical Systems (CPS). Topic: Compositional Synthesis of Finite MDPs for Networks of Stochastic Systems.

Honors and Awards


AMYTISS: PArallel AutoMated Controller SYnthesis for Large-Scale STochastIc SystemS.

AMYTISS is an advanced software tool developed in C++/OpenCL that provides parallel automated controller synthesis for large-scale discrete-time stochastic control systems which is absolutely crucial in many safety-critical applications such as autonomous driving. This tool allows to:

  • build finite Markov decision processes (MDPs) as finite abstractions of given original stochastic systems;

  • synthesize automated controllers for the constructed finite MDPs satisfying some high-level specifications (safety, reachability & reach-avoid).

AMYTISS enjoys high-performance computing (HPC) platforms together with cloud-computing services to mitigate the problem of state-explosion which is always the case in analyzing large-scale stochastic systems. This tool significantly improves performances w.r.t. computation time and memory usage by parallel execution in different heterogeneous computing platforms including CPUs, GPUs and hardware accelerators (HWAs).



Journal Papers

Book Chapters

Conference Papers