This was our

The TUfast Driverless Team has been developing a fully automated vehicle since last year that will take part in an international competition at the Hockenheimring in August. Within the scope of this talk, we will first give you general insights into the development of an autonomous Formula Student Car and give an overview on which components are necessary on the way from the conventional vehicle to a fully autonomous racing car. In the second part of the talk, we present the sensor system of the car in detail, combined with an introduction to the use of deep learning algorithms for image segmentation for recognizing the trafficable route in real time. In addition to the algorithms used in the end, we will also introduce the relevant approaches used in practice for the acquisition of annotated data with the use of classical image processing techniques as well as weakly supervised learning.

The labour market is undergoing tremendous disruptions due to advances in technology, particularly in the fields of robotics and artificial intelligence (AI). We are witnessing a perfect storm of advances in machine learning, access to large amounts of data thanks to increasingly digital transactions and cheap sensors, and the availability of faster computing power (GPUs). The main difference compared to the past is the speed of this technological change, which is now effecting every industry and increasingly impacting not only blue but also white collar (non-manual) jobs. The key questions this talk addresses: Are we still going to work in the future? How can individuals, organizations and countries prepare for a decade of job disruptions? And finally, if AI takes your job, will it find you a new one?

One fundamental issue of autonomous robots in task domains is its capability to learn new skills and to re-use past experiences under different situations as efficient, intuitive and reliable as possible. A promising mechanism to achieve that is via learning from demonstrations or observations. This talk presents a method that includes a new learning technique based on semantic representations that consider the re-usability and the inclusion of new skills in a robust manner. Then, the obtained semantics extract the essence of the observed activities in terms of human motions and object properties. The introduced method has been assessed on different Humanoid robots using different perceptual modalities, under different constraints and in several scenarios ranging from making a sandwich to driving a car. Each of the studied scenarios poses distinct and challenging levels of complexity to demonstrate, that our method does not depend on the analyzed task, thus presenting a major benefit compared to classical reasoner approaches. In this talk, the team Alle@Home will be introduced. This student team from the Technical University of Munich will participate in the RoboCup@Home international competition, where robots are expected to interact in complex and real home environments. The goal of this competition is to advance the developments of intelligent robots for future personal domestic applications.

The talk will pose some urgent questions regarding AI, such as: What is the state of the art in AI, at least what is publicly known? What is the shape of the available tools and how can organizations use them? What are some of the operating problems of employing AI and which methodological risks are known? Are data worth more or algorithms? How will future AI-aided organizations look like? The talk will attempt to give partial answers to such questions.

Poker has been a challenge problem in AI and game theory for decades. As a game of imperfect information it involves obstacles not present in games like chess and Go, and requires totally different techniques. No program had been able to beat top players in large poker games. Until now! In January 2017, our AI, Libratus, beat a team of four top specialist pros in heads-up no-limit Texas hold’em, which has 10^161 decision points. Libratus is powered by new algorithms in each of its three modules: 1) computing approximate Nash equilibrium strategies before the event, 2) endgame solving during play, and 3) fixing its own strategy to play even closer to equilibrium based on what holes the opponents have been able to exploit. The algorithms are domain-independent and have applicability to a variety of imperfect-information games such as negotiation, business strategy, strategic pricing, finance, cybersecurity, auctions, and steering biological adaptation and evolution for medical treatment planning. This is joint work with my PhD student Noam Brown.

The goal to build an AI Kicker Bot was conceived after watching a video featuring an automated foosball table that competes against human opponents. This plan was complemented with the goal to use machine learning instead of implementing game logic. This approach was already successfully used to play video games. Curiosity and the annual company’s technical retreat were reason enough to kick off this fun project that strives to bring together both the machine learning and robotics. The result after the retreat is a working prototype that uses 8 servo motors to move the bars and a commodity web cam as input source. Join us to see how this first prototype came into existence within a short amount of time, where we are now, and where we hope to go from here.

In the recent advances in approaching Nash equilibria in large imperfect information games, counterfactual regret minimization through regret matching have played a major role. This is mainly due to the simplicity of the calculations involved, while the rate of convergence in terms of iterations is rather slow. In the process of reproducing the algorithm, some new ideas have been developed at TNG Technology Consulting which significantly improve the convergence of the strategy in small games as matrix-from games, leduc hold'em and some extensions. This talk gives a brief introduction to the concept of regret matching, as well as presenting the latest results of the research that has taken place at TNG.

It is said that there are more possibilities in chess than there are atoms in the universe. However, size may not be all that matters. The different rules according to which the different chess pieces move, impose a highly nontrivial structure on the game's configuration space. Clearly, direct sampling of even small portions of this space is out of reach. Yet, the task of figuring out properties of a state space that is too vast to enumerate, is a familiar one for statistical physics. In this talk I will show how we applied transition-path sampling, an advanced Monte-Carlo simulation method that is usually used to study crystallization, protein folding or the flow behavior of polymers, to chess. The simulations show that chess' state space decomposes into a large number of weakly connected "pockets" that reflect the pawn structures emphasized by good chess players.

Chess is alive and well; as you read this, tens of thousands of people are playing it online. Despite the large demand, the market of online chess servers is saturated, and the competition is fierce. Nowadays, most chess servers are traditional businesses, relying on ads and/or paywalls. But there exists an exception. lichess.org is a free-software, free-service, and free-of-ads chess server. It makes the hypothesis that a good service can be built by a community of volunteers, out of passion for the game, and with humanist values at its core. What happens when users are treated as contributors rather than clients? How sound is it to open the source code of an online gaming server? Can a community run its own free service and compete with big companies? Turns out it can, and it's a lot of fun.

Big data management is becoming more and more important in medicine, especially in the field of tumor medicine (oncology), because new investigation procedures generate strongly increasing data sets, notably, genome data (next generation sequencing), but also image, sensor, clinical and self-reported data from the patient (Quantified Self). Particularly in oncology, diagnostic and therapeutic procedures have been improved by new IT procedures. Due to the growing amount of data, there is a need for clinical decision support. However, the integration of data sources for research and patient care remains a major challenge because the data models in medicine are still not standardized. Patient data are particularly worthy of protection (data protection) and IT systems in medicine require particularly high reliability requirements. The lecture gives an overview of the possibilities and limits of big data procedures in medicine based on selected current examples.

Witnessing the rise of stream processing from the driving seat, we see Apache Flink® and associated technologies used for a wide variety of business applications, from routing data through systems, serving as a backbone for real-time analytics on live data using SQL, detecting credit card fraud, to implementing complete end-to-end social networks. Such applications enable modern data-driven businesses where decisions and actions happen in real-time, and transform traditional businesses to become more data-driven. Observing the variety of these applications implemented using Flink, it becomes apparent that the traditional dividing line between analytics and operational applications is becoming more and more blurry. Historically, operational applications were built using transactional databases, and analytics were done offline. In contrast, Flink’s, state, checkpoints, and time management are the core building blocks for both operational applications with strong data consistency needs, and for real-time analytics with correctness guarantees. With these shared building blocks, developers start building what is arguably a new class of data-driven applications: applications that are operational in that they serve live systems and at the same time analytical in that they perform complex data analysis. Following application architectures like CQRS and using new features like Flink’s queryable state, streaming analytics and online applications move even closer to each other. In this talk, guided by real-world use cases, we present how the unique core concepts behind Flink simplify the development, deployment, and management of data-driven applications, and we conclude with a vision for the future for Flink and stream processing.

At Xero, we are on a mission to find ways to make our accounting software smarter. In addition to seeing the data that you entered presented back to you in a clean, beautiful format, our algorithms can use what they know to spare you entering it in the first place. Our technology does not just collect information: it learns from what it stores. Ultimately this is turning the boring and the mundane into an automated process - giving our customers more time to focus on what’s important for them. The challenge of automating out the boring bits of accounting is not in the algorithms themselves: it's how we engineer our back end to produce the right information at the right time, and in some cases the challenge is identifying what that right information even is! Come along and hear how we use machine learning pragmatically to save time for our customers.

A growing proportion of human activities such as social interactions, entertainment, shopping, and gathering information are now mediated by digital devices and services. Such digitally mediated activities can be easily recorded, offering an unprecedented opportunity to study and measure intimate psycho-demographic traits using actual―rather than self-reported―behavior. Our research shows that digital records of behavior, such as samples of text, Tweets, Facebook Likes, or web-browsing logs, can be used to accurately measure a wide range of traits including personality, intelligence, and political views. Such Big Data assessment has a number of advantages: it does not require participants’ active involvement; it can be easily and inexpensively applied to large populations; and it is relatively immune to cheating or misrepresentation. If used ethically, it could revolutionize psychological assessment, marketing, recruitment, insurance, and many other industries. In the wrong hands, however, such methods pose significant privacy risks. In this talk, we will discuss how to reap the benefits of Big Data assessment while avoiding the pitfalls.

What would happen if one could communicate with holographic augmented reality spectacles with far away acquaintances as if they were in the same room? The hardware hacking team of TNG Technology Consulting GmbH has developed a software prototype for holographic telepresence for Microsoft HoloLens. Thus, similar to the Star Wars series, it is possible to project a person as a hologram into their own field of view. The idea of augmented reality (Extended Reality) has existed since the late 1960s. With the HoloLens, Microsoft has succeeded in bringing a self-sufficient device onto the market, which expands the real world immersively with computer-generated elements and so-called holograms. Within the framework of the lecture, ideas and concepts in the field of mixed reality are going to be introduced and explained. Technical details of the HoloLens are presented and demonstrated by means of several interactive live demonstrations. In addition, simple programming examples for the HoloLens using Unity 3D are developed in live coding.

Birdly fulfills people's ancient dream of flying by deploying the latest virtual reality (VR) and advanced simulator technology. The installation creates a vivid full-body experience that instantly lets you forget the mechanics and computer codes behind this spectacular apparatus. The immersive and interactive nature of the installation serves one goal: to enjoy the ultimate freedom of a bird. The experience allows you to intuitively explore the skies, cities like New York and any realistic or imaginary territory that can be built with 3D computer graphic tools.

Hyperloop is a high-speed transportation concept proposed by the Tesla and SpaceX founder Elon Musk in 2013. It envisions transportation capsules levitating inside a low-pressure tube close to the speed of sound. By drastically reducing aerodynamic drag in the aforementioned evacuated tube and reducing surface friction by i.e. magnetic levitation or air-bearings, the concept becomes technically feasible. To foster interest and foster innovation, SpaceX has organized an international competition to test sub-scale versions of Hyperloop vehicles on a one mile purpouse-build tube in Hawthorne, California. The "WARR Hyperloop" student team from the TU Munich successfully demonstrated their prototype in the competition with over 700 entries and 30 finalists, and secured the grand prize for the "fastest pod" in the end. Complementary to last year's presentation at BTD9, the design and build process will be reviewed in hindsight of the unique challenges posed by Hyperloop concept and the competition constrains, before eventually taking a look back at the competition week, the final run and concluding why "wheels are pretty good".

It's been almost 60 years since the introduction of the LEGO® building system, and it has evolved significantly since then. Originally started as simple building blocks for young children, it has grown to include electric motors, gear wheels, pneumatic components, and even programmable computers - all of which are now used by adults as well. Paul will discuss how the best LEGO® model builders in the world use toy parts to create complex mechanisms, how new technologies such as the 3D printing are affecting this hobby, and how far we can currently push the envelope by using LEGO pieces for tasks they were never intended for. He will demonstrate multiple examples of real-world engineering recreated using parts from LEGO sets, several examples of complete, custom working motorized models, and he will talk about various challenges and opportunities that model builders usually face.

Most people wouldn't imagine that the world's fastest radio controlled airplane has no propeller, jet, or any source of propulsion on board. Dynamic Soaring is a unique method of soaring which has enabled radio controlled gliders to achieve speeds in excess of 835kph.  We will explore the fundamentals and progress of dynamic soaring and examine the challenges associated with designing, building, and flying an un-powered model airplane at speeds comparable to modern passenger jets.  We will also give some thought to practical applications of dynamic soaring and discuss where it could be applied in the future.

The TUfast Eco Team, which is part of the student initiative TUfast founded in 2002, offers motivated students from the TU Munich a chance to gain practical experience in the development of highly efficient vehicles and to actively deal with the latest advances in electromobility. The focus is on the development of innovative solutions through interdisciplinary cooperation. Within the scope of this talk, we will first give you a brief overview of the TUfast Eco Team and then have a closer look at the development process and the technical background of our current vehicle. In contrast to the years before, we have not concentrated on the prototype class this season, but have switched to a completely new vehicle class - the Urban Concept class of the Shell Eco-marathon. An urban concept differs in several respects from the prototype class, since the main focus is now not only on the lowest possible energy consumption - a limited roadability is now additionally required. Basically, it is about developing a highly efficient vehicle, which is closer adapted to the everyday conditions and requirements. This means, among other things, an upright sitting position, higher weather resistance, a complete outdoor lighting system and much more.

A Live-Hacking-Show, that so far inspired thousands of people worldwide. This talk will bring you closer to the world of hacking and allows you to take a look at the poison cabinet of IT. In the process, security gaps in computers and mobile phones, which affect us all, are going to be uncovered in an entertaining way. This talk means infotainment: It is understandable for everyone and does not require any technical know-how. Thus, everyone will detect fascinating everyday examples throughout this lecture. Passwords will be cracked within seconds; discrediting information will be uncovered and without further ado a smartphone will be hacked. Every example is live and real, but of course anonymous. Every listener will be able to recognize himself – no one is going to be exposed. For risks and side effects please consult your data protectors or Tobias Schrödel.

Nearly every new development of software involves the use of open source components, generally covered by different licenses. Because a violation of licensing terms leads to a copyright infringement, an elaborated license management for open source is necessary. The new Version 1.1 of the OpenChain standard for the open source license compliance in the supply chain was introduced April 2017. The lecture explains the structure and content of the specification of the standard and how the use of open source software can be legally protected.

This talk will deal with the topic of data protection with regard to the new Federal Law for Data Protection. In particular, a closer look will be given on how seriously personal data is taken, which new sanctions were enacted, as well as the impact of the changes on compliance issues. In addition, this talk deals with examples comparing the situation before and after and describes vividly what is really no longer possible and what challenges arise for the practice.

Fuzzing is a technique in semi-automated security analysis which traditionally feeds random data to the inputs of a computer program. The goal is to find inputs which trigger some unexpected behavior of the program, pointing to weaknesses in the implementation and therefore revealing unknown security problems of the program. Recently, there were considerable advances in this field which enable us for the first time to fuzz system-call interfaces of any pre-compiled kernel by constructing a suitable user-land driver, which connects any fuzzer to the kernel of interest. We apply this idea to the L4 microkernel L4Re and explain the techniques to analyze the implementations by examining for example the IPC interface in the L4Re microkernel. Furthermore, we elaborate on the different fuzzers available today and present the results of our analysis.

Functional Programming is gaining a lot of excitement. While this style of programming has been around for a while, the mainstream languages have finally and recently caught up. Programming in functional style is a lot less about the syntax and a lot more about the mindset. What does it mean to program in functional style? More important, why should we care? That's the focus of this presentation.

In our information-driven world it is easy to forget that technological progress has always crucially depended on progress in materials. The development is still too much based on trial and error, slow and expensive, and predictive computations of materials would be desirable (“Dirac’s challenge”). This problem of quantum mechanics is still not satisfactorily solved, in particular in the important field of “correlated materials” (which I will explain) comprising high-temperature superconductors, rare earths and many other technologically relevant material classes, where the standard workhorse “density functional theory” fails. I will review the current state of the art and show that success for these difficult materials can now be achieved. Nevertheless, further progress in methods is needed, and I will show the benefits of combining traditional methods from solid state physics with methods based on quantum information theory.

Scrum, that won’t work!” - Several mechanical and plant engineers are skeptical regarding the use of modern concepts for and methods of software development. In many cases, software development is also not their core competency. Indeed, the product development process usually follows a classic waterfall and factory approach. Syskron GmbH has started in the fall of 2016 an exciting restructuring process. The aim is to evolve its working procedures using new concepts and approaches and, thereby, modernize the MES software products. This field report shall illustrate the clash between two mindsets, the constrains and challenges faced, and the solutions found to counteract them. But above all, in this talk you will find out what has already been successfully implemented, despite all obstacles, and what has been accomplished since.

The world is turning faster and faster. We live agile with Scrum, Kanban, Continuous Integration/Deployment, and DevOps; we build prototypes, MVPs and try new technologies. Nevertheless, we need structures and procedures that are more fundamental and more sustainable. In our fast-moving daily routine of our systems and products, does software architecture get the necessary attention? Do we test it as we today (luckily almost self-evident) write unit tests? How do we document it? How do we sustain it? What are the organizational framework conditions? What do we have to do to stay master of our architecture? The talk examines this area of conflict and reflects corresponding experiences.

The project Check_MK, launched in 2008, developed within only a few years from a small collection of scripts to a mature IT monitoring system, which is becoming increasingly popular and whose further development in the meantime keeps a company with 15 employees occupied. Part of the success is based on some very unusual design decisions in the software architecture. Mathias Kettner, the founder of the Check_MK project, gives insights into the inner life of Check_MK after a short live demo of the software and shows by some examples where Check_MK is "different " and why the courage to go own ways in software development can pay off.

Test infrastructure is an important part of software development projects. Fast test-feedback cycles allow developers to commit more often, to move fast and resolve problems quickly. Test runtime is often dominated by integration tests, in some projects running for hours. Databases are often shared resources or require locking. The first step to faster tests is therefore to encapsulate the whole test system into containers, so that previous points of contention can be run in parallel. Currently, test infrastructure is often run in-house for convenience, frequently leading to a trade-off between server cost and test speed since the usage outside office hours is limited. We show that by transferring the test system to an automatically scaling kubernetes cluster, we can gain significant speedups and their associated benefits.

Data Science is to software development what alchemy is to technical construction. They use the same basic tools and methods and exist alongside each other but the former puts its focus on fast prototyping and trying of new methods rather than targeting stability and performance. In my talk, I will use development requirements and applications from the Mobility Data Services Center as an example to show how –and why- we use a python-based DevOps approach to tackle the Data Analytics requirements of the rail world.

At Facebook we built and open-sourced the Haxl framework because we wanted a way for engineers to be able to exploit concurrency without thinking about it. In this talk I'll illustrate how Haxl works and explain some of the ways it's been successful at helping engineers worry less about performance, enabling them to focus more on correctness. I'll also talk about some forthcoming improvements to Haxl that significantly generalise the way the framework handles concurrency, and demonstrate some compelling use cases.

Restarting the Browsers wars. After two browser wars (Netscape vs. Explorer and Chrome vs. Firefox), Cliqz is starting the browser war III with a focus on search & privacy as the core of the browser. We will address the questions: Why Now? Why from Germany? and Why with Burda? does that work? In the 3rd question, I will address the successful digitization of Burda since 1994 where I joined.

Zalando has introduced Radical Agility two years ago: an organisational concept that relies on team autonomy. This raises the question of how architectural work is realized in such an environment. I will illustrate how Zalando SE ensures an uniform architectural image of the services with architectural principles and guilds. Thereby I discuss organizational peculiarities and the main guidelines which we use in our development processes and furthermore I give you a brief insight into exciting technical challenges we face.

During this talk we will cover what Ceph is exactly, what it does, and why you need it. To do so, Mr. Emmerich provides details of his last 4+ years of experience with Ceph as well as of his learning curve. Her further compares the current solutions that exist on the market. Afterwards, he shows live a demonstration installation of a complete cluster, using the explained software. There is also enough time left for a Q&A.