New deep learning & big data frameworks application to simultaneous classification, segmentation and pose detection for future self-driving cars

New deep learning & big data frameworks application to simultaneous classification, segmentation and pose detection for future self-driving cars

Our project assumes that the launch of automated vehicles on public roads will only be successful if a user centric approach is used where the technical aspects go hand in hand in compliance with societal values, user acceptance, behavioural intentions, road safety, social, economic, legal and ethical considerations.
This approach must be understood as a multidisciplinary task, including engineering and technology, but also social sciences. Based on our previous and recent results on both, intelligent predictive methods for VRUs protection (IMPROVE project 2014-2017) and ITS-G5A V2X communications for cooperative automated driving (best team with fully automation at the Grand Cooperative Driving Challenge 2016), the main goal of this proposal is the application of new learning & big data frameworks to simultaneous classification, segmentation and pose detection for future self-driving cars. These will be the starting point for the integration of predictive techniques to make self-driving vehicles closer to humans when making critical decisions. Thus, this proposal aims at enhancing ADAS through a robust system that will combine predictive approaches and adaptive HMIs to create a consistent and seamless support of the driver of automated vehicles towards predictive intelligent, efficient, and safe interaction between automated vehicles and VRUs, including cyclists. New predictive models will based on other vehicles’ measured and communicated variables, VRUs perceived kinematics and environmental context awareness, including road context and a priori information. A priori information will be incorporated by means of probabilistic and deep learning techniques that will allow obtaining the most likely intentions of other drivers and VRUs in the road scene, thus contributing to more realistic traffic assumptions and more efficient and safer ADAS. The adoption of ITS-G5A V2X communications will add flexibility and robustness to the predictive system for anticipating other road users’ intentions, allowing the development of potentially standardized solutions. Adaptive HMI strategies will be studied by applying a user-centric approach. Repeatable and measurable tests in relevant environments will be performed to both assist the driver with automation when needed and to achieve smooth transitions between automatic and manual driving. New adapted and equipped tests will be delivered for the new automated vehicles, aligned with proposals for human behavior safety testing and safety standards. This proposal aims at paving the way for the further adoption of automated vehicles by advancing on validation protocols. In this way, a set of new Euro NCAP type scenarios for automated driving in presence of VRUs and other vehicles will be implemented using the prediction and V2X communications frameworks, performing a repetitive analysis using different drivers and VRUs, with special focus on user-acceptance. The adaptive HMI concept should be applicable to the defined Euro NCAP and automated driving testing use cases and scenarios, supporting the drivers and the VRUs in order to complete the tasks and transitions in a safe and acceptable manner.