In the run-up to the ScaleUp 360° Auto Vision, we.CONECT spoke with Tudor Nicosevici, Head of ADAS Advanced Engineering Romania at Continental, about understanding the true complexities of computer vision and sensor fusion technologies.
Tudor Nicosevici: As head of ADAS Advanced Engineering, Continental Romania, my role is to coordinate my team’s activities from the technical and managerial point of view. The team is responsible of pre-development activities for assisted and autonomous driving systems. In particular, our area of covers topics such as: Artificial Intelligence, sensor fusion, ego motion, localization, and environment modeling.
Tudor Nicosevici: Autonomous driving brings multiple significant advantages. When talking about autonomous driving, convenience is probably the most common one that come to one’s mind. However, the biggest motivating factor for me is substantial improvements in safety that autonomous driving is able to provide. Moreover, combining autonomy with connectivity and cooperativity between vehicles, can lead to major traffic optimization, which ultimately reduces pollution and time we waste during commutes.
Tudor Nicosevici: In my opinion, we are still far away from large-scale, commercial-level L5 autonomous driving. In the last period, the expectations of the industry with respect to L5 were dialed back, and with good reason: as we pushed towards autonomous driving, we started to understand the full scale of the challenges that we have yet to overcome. These challenges mainly stem from the complexity and dynamicity of the transportation environment combined with the high safety standards of the automotive industry.
Tudor Nicosevici: Cameras and computer vision systems play a pivotal role in autonomous driving, as they are the single richest source of spatial and semantic environment information. However, due to their limitations in accurately measuring distances and their sensitivity to certain lighting and weather conditions, complementary sensors such as lidars and radars are paramount in order to ensure safety of autonomous systems. In this context, reliable sensor fusion techniques are required that take into account the inherent limitations and advantages of each sensor and the specific environmental conditions.
Tudor Nicosevici: Significant improvements in autonomous driving will depend on our ability to develop reliable, scalable, understandable and safe AI systems. For this, we need better machine learning architectures and improved training and validation of such systems. Another crucial aspect is data. Here, we need a more collaborative mindset across the industry, as it is very difficult that a single company can obtain sufficient data to train reliable AI systems.
Tudor Nicosevici: AI is showing great potential in perceiving and modeling complex scenarios, is paving the way for automated and autonomous driving. However, current AI algorithms have a number of drawbacks that make their applications limited in safety critical systems, as they are largely “black boxes” and cannot handle probabilistic models. In this talk, we discuss Bayesian Deep Learning, a novel approach which promises to alleviate these shortcomings, while providing better means to develop and validate autonomous and automated driving systems.
About Tudor Nicosevici:
Tudor Nicosevici is the head of ADAS Advanced Engineering Romania, Continental Automotive. Additionally, he is responsible for the technical coordination of research and development in vehicle ego-localization and environment modeling within the Advanced Engineering group. His areas of expertise are computer vision, robotics and machine learning, with a M.Sc. in sensor fusion and a PhD in robotics and computer vision. Prior to joining Continental he coordinated activities of technology transfer between the academia and the industry sector.
Join Tudor Nicosevici’s webinar session at 2019 ScaleUp 360° Auto Vision:
Session title: Reliable data fusion in autonomous driving by means of statistical modelling of deep learning systems
Date: 18 October 2019, 09 AM (CEST)
Follow this link to sign up now for free: Online registration
Signing up automatically gives you access to all webinar sessions at ScaleUp 360° Auto Vision, 17 - 17 October 2019
ScaleUp 360° Auto Vision is the digital event deep diving into Computer Vision and Sensor Fusion technologies for autonomous vehicle perception. Experience 2 days with 12 live webinars and case studies by stakeholders involved in computer vision, sensor hardware, image processing and sensor fusion in the Level 5 automation scene. Learn, engage and discuss automotive tech innovation in real-time with thought leaders across the globe – right from your desk.