Model Predictive Control for Motion Planning in Urban Environments
Every year more than 20 million people are involved in road accidents, mostly caused by human errors. According to the World Health Organization, approximately 1.3 million people lost their lives in these accidents. Half of the victims are vulnerable road users (VRUs), such as pedestrians and cyclists. Self-driving vehicles can help reduce these fatalities. This talk presents a VRUs-aware local motion planner based on model predictive control (MPC). Our planner strongly relies on the interaction with the environmental perception for navigation in complex urban environments. The perception module detects and estimates the paths of the VRUs over a prediction horizon, while the planning module exploits these paths to plan collision-free trajectories. Real-life experiments shows the potential of our design for the future of urban driving.
Laura Ferranti received her Ph.D. from Delft University of Technology (TU Delft), Delft, The Netherlands, in 2017. She is currently a postdoctoral researcher in TU Delft. Her research interests include: optimization and optimal control, model predictive control, embedded optimization-based control with application in flight control, automotive, maritime transportation, and robotics. The main techniques involved in her research are proximal gradient methods (deterministic and stochastic versions) and splitting methods (such as, the alternating minimization algorithm and the alternating direction method of multipliers) for convex and nonconvex (nonlinear) optimization.
Planning via Constrained Markov Decision Processes
In this talk I will retrace our research journey in the area of autonomous and intelligent systems. I will then present a high-level architecture for autonomy and dive into the technical details of a planning system we have implemented to bridge the gap between mission and motion level planning. In particular, I will discuss the problem of planning in a domain where the outcome of an action is probabilistic. Such planning problems are typically solved using Markov Decision Processes (MDP). To account for realistic mission requirements, we extend the MDP planning framework by accepting behavioral goals described in Linear Temporal Logic, and by incorporating constraints on critical resources or probability of mission success. The resulting planning problem can be solved via Constrained Markov Decision Processes. I will conclude the talk with a list of challenges that we consider still open in the area of autonomy for aerospace applications including explainability and assurance.
Dr. Pinto is a Technical Fellow at the United Technologies Research Center (UTRC). He works at the intersection of embedded system design, model-based design automation, and autonomous and intelligent systems. His current interests include architectural design for autonomous systems, knowledge representation, algorithms for high-level reasoning and decision making, safety assurance for autonomous systems, and compositional design methodologies. He is the recipient of the 2014 UTC Outstanding Achievement Award for his contributions in the area of Autonomous Rotorcraft Technology Development and Demonstration, and of the 2016 UTRC Technical Excellence Award. Dr. Pinto earned a Laurea degree in Electrical Engineering from the University of Rome “La Sapienza”, and a Ph.D. degree in Electrical Engineering and Computer Sciences from the University of California, Berkeley.
Software Exploitation: Hardware is the New Black
What would the world be like if software had no bugs? Software systems would be impenetrable and our data shielded from prying eyes? Not quite. In this talk, I will present evidence that reliable attacks targeting even "perfect" software are a realistic threat. Such attacks exploit properties of modern hardware to completely subvert a system, even in absence of software or configuration bugs. To substantiate this claim, I will illustrate practical attacks in real-world systems settings, such as browsers, clouds, and mobile. The implications are worrisome. Even bug-free (say formally verified) software can be successfully targeted by a relatively low-effort attacker. Moreover, state-of-the-art security defenses, which have proven useful to raise the bar against traditional software exploitation techniques, are completely ineffective against such attacks. It is time to revisit our assumptions on realistic adversarial models and investigate defenses that consider threats in the entire hardware/software stack. Pandora's box has been opened.
Cristiano Giuffrida is a Tenured Assistant Professor in the Computer Science Department at the Vrije Universiteit Amsterdam. His research interests span across several aspects of computer systems, with a strong focus on systems security. He received a Ph.D. cum laude from the Vrije Universiteit Amsterdam under the supervision of Andy Tanenbaum in 2014. He was awarded the Roger Needham Award at EuroSys and the Dennis M. Ritchie Award at SOSP for the best PhD dissertation in Computer Systems in 2015 (Europe and worldwide). He was awarded a VENI grant (the Dutch Equivalent of a NSF CAREER Award, PhD+3) in 2017. He has served on the program committee of a number of top systems and security venues, such as SOSP, OSDI, EuroSys, S&P, CCS, NDSS, and USENIX Security.
Bridging the Gap Between Requirements and Model Analysis: Evaluation on Ten Cyber-Physical Challenge Problems
We present a framework for introducing high-level requirement specifications in the automated analysis of dataflow models. By integrating the Formal Requirements Elicitation Tool (FRET) with the CoCoSim analyzer, our framework enables the analysis of hierarchical Simulink models against requirements written in a restricted English language. More precisely, we support: automatic extraction of Simulink model information and association of high-level requirements with target model signals and components; translation of temporal logic formulas into synchronous dataflow specifications as well as Simulink monitors; and interpretation of counterexamples produced by the analysis both at the requirement and model levels. For the analysis, we use the Kind2, Zustre, and Simulink Design Verifier (SLDV) tools. The features provided by our framework are NA generic and can be used to integrate other requirements elicitation and Simulink/Lustre analysis tools. We report on lessons learned from the application of our approach to the Lockheed Martin Cyber-Physical, aerospace-inspired challenge problems.
Anastasia Mavridou is a Computer Scientist at the Robust Software Engineering group at NASA Ames. Previously, she was a PostDoctoral Researcher at the Institute for Software Integrated Systems at Vanderbilt University. She received her PhD in 2016 from Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland. Her research interests revolve around formal modeling and analysis of systems with a focus on correct-by-construction techniques.