Wolfgang Hillen Summer School 2022

Prof. Dr. Andreas Herkersdorf
Technische Universität München
Fakultät für Elektro-/Informationstechnik - Lehrstuhl für Integrierte Systeme

E-Mail: herkersdorf@tum.de


Dr. Fadi Kurdahi
CECS - UC Irvine
The Henry Samueli School of Engineering
University of California,
Irvine CA 92697-2625, U.S.A.

e-mail: kurdahi@uci.edu


Modeling and Design of Next Generation Self-X MPSoC Platforms

17. – 21. October 2022 – UC Irvine, California

With the recent successful developments in artificial intelligence, several research activities are being carried out to bring self-X properties (as self-adaptivity, self-awareness, self-organization, self-healing, etc.) to embedded systems, leading the path toward smart embedded systems. One key step is to abstract the complexity of enabling self-X properties into a model allowing the control and steering of a system’s dynamics at runtime, to enable these self-X features.

Research groups of the University of California, Irvine (UCI), Technical University of Munich (TUM) and Technische Universität Braunschweig have proposed a model called the ‘Information Processing Factory’ (IPF) that aims to show how these self-X properties can be achieved across multiple abstraction levels of a Multi-Processor System-on-Chip (MPSoC) platform and for a mixed critical application scenario. Based on a platform-centric design approach, IPF combines self-organized learning and formal reactive methods to guarantee the applicability of such self-X systems in safety-critical and high-availability applications.

The Wolfgang Hillen Summer School by UCI and TUM wants to share the finding of IPF in an interactive and practical manner with three goals in mind:

The Wolfgang Hillen Summer School takes place at UCI with experts from local technology companies to present the State of the Art of self-X systems in industry and researchers from University of California, Irvine, Technical University of Munich and Technische Universität Braunschweig conveying the IPF concept and obtained results in form of seminars and workshops.

More information at:
https://www.ei.tum.de/lis/studentische-arbeiten/wolfgang-hillen-summer-school-2022/.

 

Final report

Before our Wolfgang Hillen Summer School on „Modeling and Design of Next Generation Self-X MPSoC Platforms“ could take place at UC Irvine in fall 2022, we had to experience two years of throwbacks due to the COVID pandemic and travel restrictions. Also in this context, we would like to thank BaCaTeC, and especially Dr. Rosenzweig, for his continued support in making the postponement possible without further efforts.

The Summer School was held from October 17 to 21, 2022 in Irvine, California. For five days (and evenings), master students and doctoral researchers from Germany (Munich and Brunswick) as well as from California (Irvine) explored a variety of research topics in the context of interconnected Self-X MPSoCs. On the first day, people were inspired by keynotes from industry and academia, followed by three hands-on workshops starting from the second day.

After a warm welcome and a short game to introduce all participants to engage the exchange between persons from the three different locations, we had a series of talks on the first day. The keynotes conducted by representatives of PerceptIn, Mercedes Benz, NVIDIA, UCI, SDSU, TUBS and TUM covered safety and verification, Machine Learning (ML), self-awareness of systems, networking and modeling. To keep the discussions about these topics ongoing, this first day was completed by a BBQ.

The following three days were allocated to hands-on workshops, thereby the students could choose one out of three different topics for the duration of the Summer School. One workshop focused on Time-Sensitive Networking. There, the students were first given the task of building a representation of a future vehicle network in a zonal structure (cf. Figure 1) using IDA’s OMNeT++-based TSN Simulator. In doing so, they received inputs in the methodical advantages and challenges of using simulation tools for developing and testing complex systems, networks and protocols as well as first hands-on experiences with the technology itself. Incidentally, the student stumbled over typical challenges of simulation tools, namely different abstraction levels that have to be considered when building a simulation. Similarly, using all the (TSN) parameters given by the TSN Simulator proved challenging, the students were given a glimpse into the complexity of those systems. In the end however, they successfully managed setting up a network simulation that transmits (dummy) “sensor” data from multiple sensor groups to a central sensor fusion node and running first experiments. The second task tackled dynamic scenarios that require traffic optimization and resource management, and thereby primarily addressed scientific issues, providing a more detailed look into the future research directions that significantly advances existing technologies in that domain. The students were tasked with integrating an external V2X sensor data stream (e.g., as coming from a different platoon member or the infrastructure) into the already stressed network. Quickly they discovered that simply adding the stream on top results in packet drops across many streams – a situation that is not viable. Hence, they had to coordinate the streams using a resource manager and apply a novel object caching strategy that optimizes the data transmitted itself and serves as an alternation distribution mechanism, a trait enabled by the data-centric approach to the communication architecture. The experiments performed by the student showed that the resource management in combination with the ‘smart’ caching allowed for integrating the new traffic stream without violating any timing and safety constraints.

Figure 1: Future Zonal Vehicle Network

In another workshop, the students were exposed to building a platooning platform. The workshop primarily consisted of manufacturing truck heads, truck trailers, high-level controllers, and low-level controllers. The scaled truck manufactured by Tamiya has precision and components reminiscent of a real truck. The students completed one truck by faithfully fulfilling their roles and demonstrating its operation. The students moved to the next step with an understanding of the entire truck configuration. The high-level controller runs the ROS operating system, which connects various machine-learning algorithms and low-level bridges to run the entire system. The entire system will be the basis for future operational tasks and projects of the Information Processing Factory project.

Our third workshop focused on lightweight ML on FPGAs. Therefore, the first task for the students was to familiarize with rule-based reinforcement learning. Towards this goal, a Matlab simulation environment was provided, where the students investigated the effect of different parameters and their importance. At the end of the day, the students were able to balance an inverted pendulum. The second day of the workshop was dedicated to investigating possible optimizations of the algorithm to transfer it on the FPGA. The students were given a straight forward VHDL implementation of our rule-based RL engine. Here, the drawback of sensor values with fine granularity, floating point numbers in HW and certain potential algorithmic modifications were found. In a next step, optimizations proposed by the students (e.g., quantization of measurement values) were tested in the Matlab simulation. Finally, on the third day of the workshop the insights gained the days before were transferred into VHDL code. Here the students could finally demonstrate their coding skills. Although there was a tight timeline for this workshop, we never missed any opportunity to discuss how the research and knowledge of the participates could be incorporated into the setup to further improve or add performance to the design.

The last day of the Wolfgang Hillen Summer School was spent by exchanging all gained experiences in form of presentations and the hand-over of the participants’ certificates.

All these activities were accompanied by social events in form of the daily lunch, walks on the beach in the evening and dinner with local food.

We enjoyed every moment of the summer school and the various exchanges we had together. These exchanges resulted in plenty of new ideas for the research of participants and organizers. Many thanks to BaCaTeC for their generous support to make such events and experiences possible.