Title: Automated cars - slalom use case and time-deterministic infrastructure.
R&D in automated cars represents an attractive challenge for every control engineer. This talk will present an application with experiments and optimal use of the car's infrastructure.
First, we deal with an automated car aimed at performing a slalom use case defined by Porsche Engineering. The positions of the cones are not known, they may have unequal distances, and the U-turns may be symmetric or asymmetric. Namely, we show our trajectory planning algorithm based on optimization techniques using an external layer deciding a position of the nearest waypoint and an internal layer generating the optimal trajectory. The system's validation with Porsche Panamera reveals interesting observations about the components' precision, frequency, and sensitivity.
Second, we deal with the determinism and efficiency of the car's infrastructure. We suggest time-triggered scheduling algorithms to guarantee timing requirements of safety functionalities implemented as computational tasks and communication messages. Namely, we focus on the periodic scheduling of messages on FlexRay and IEEE 802.1Qbv Time-Sensitive Networks.
Everything must be safe and fast since we desire safe and fast automated cars.
Zdenek Hanzalek obtained his Ph.D. degree in Industrial Informatics from the Universite Paul Sabatier Toulouse and a Ph.D. degree in Control Engineering from the CTU. in 1997. He worked on a parallel algorithm optimization at LAAS CNRS in Toulouse (1992 to 1997) and LAG INPG in Grenoble (1998 to 2000). Later he founded the Industrial Informatics Group at CTU in Prague, dealing with combinatorial optimization, scheduling, autonomous cars, and real-time embedded systems.
Zdenek teaches a Combinatorial Optimization course, and he is active in the scheduling community while driving the worldwide seminar schedulingseminar.com. He acquired and led many industrial contracts (e.g., Porsche, Skoda, Volkswagen, UNIS, AZD, Rockwell, Air Navigation Services, EATON, PPL). Besides this, Zdenek acted as the founder of the SW development group at Porsche Engineering Services in Prague, which has several hundreds of employees today. He is the author or co-author of 55 journal papers and approximately 70 conference papers, mainly dealing with applications of optimization algorithms in computer and manufacturing systems.
Title: Proactive Autonomous Navigation in human populated environment
Autonomous navigation in human populated environment is difficult as it is facing the freezing robot problem where generally reactive techniques fail. With a certain level of density, there is no solution if we don’t take into account the future evolution of the environment over a time horizon. There are different aspects to consider in this global problem. Human (or simply pedestrian with or without an electrical mobility devices) behaviors require to be observed or learnt in order to predict their evolution. An accurate and realistic model of such agent is necessary. In that area, recent advances have been done in by enhancing the classical Social Force Model. The second aspect concerns the observation as human represents an Hidden dimension and the question is what is necessary and can be observed? The third aspect deals with the control in order to monitor the action of the robot. The question is: what is the best action to do taking into account the knowledge we have and the observation we do in order to join a particular place in the human populated environment? During the presentation, I will present the recent advances that we have done in different research projects.
Philippe Martinet graduated from the CUST, Clermont- Ferrand, France, in 1985 and received the Ph.D. degree in electronics science from the Blaise Pascal University, France, in 1987. From 1990 to 2000, he was assistant Professor with CUST. From 2000 until 2011, he has been a Professor with Institut Franc ̧ais de Mécanique Avancée (IFMA), Clermont-Ferrand. In 2006, he was a visiting professor at the Sungkyunkwan university in Suwon, South Korea. In September 2011, he moves to Ecole Centrale de Nantes and LS2N. In November 2017, he moves to Inria Sophia Antipolis as Research director. His research interests include visual servoing of robots, multi-sensor-based control, force vision coupling, autonomous guided vehicles, modeling, identification and control of complex machines. From 1990, he is Author and Coauthor of over three hundred seventy references.
Title : Towards autonomous vehicles: how does AI get on board?
Armed with automotive-grade sensors and new dedicated hardware, all mobility solutions, from private cars to robotaxis or droids, can now benefit from cutting-edge AI models aboard. In this talk, we will explain how the introduction of more and more sensors and cutting edge-AI in cars over the years is leading the automotive industry in a long journey from parking assistance to driving assistance to full autonomy. We will deep dive in few challenges on this road:
· How machine perception aims at turning sensory inputs into some useful understanding of the world
· How imitation and reinforcement learning can help to build efficient AI model to predict vehicle trajectories as well as vulnerable road users
· How a smart data management and trustworthy AI approaches are important to go from research to products
Finally, we will show the results of all that on Valeo Drive4U driving on open roads with only Valeo serial production sensors.
Head of AI at Valeo Driving Assistance Research, Dr. Xavier Perrotton is leading research on artificial intelligence and autonomous driving to make smarter mobility an everyday reality. His expertise includes computer vision, AI, deep learning, applied math, software architecture, sensors and high performance computing. Prior to joining Valeo, he was research scientist, then research project manager at Airbus group innovations, working on computer vision and Augmented Reality, transforming ideas into prototypes and contributing to their industrialization. He holds a master of electrical engineering from Supélec, a master of applied Mathematics from Paul Verlaine - Metz University and acquired his Ph.D. at Telecom ParisTech.