IV2017 - IEEE Intelligent Vehicle Symposium
13/06/2017 — 13/06/2017
Redondo Beach, USA
Dreams4Cars is organising the workshop “Cognitively Inspired Intelligent Vehicles” as part of IV2017 Conference.
The Intelligent Vehicles Symposium (IV’17) is a premier forum sponsored by the IEEE Intelligent Transportation Systems Society (ITSS). Researchers, engineers, practitioners, and students, from industry, universities and government agencies are invited to present their latest work and to discuss research and applications for Intelligent Vehicles and Vehicle-Infrastructure Cooperation.
The initial focus of a research field interested in the development of autonomous systems is often on algorithms designed from a pure engineering perspective. Research on intelligent vehicles, for example, is currently in this stage. At the same time, there are other fields interested in autonomous systems that have been in existence for longer. One clear example is research on (cognitive) robotics and artificial cognitive systems. A lesson that can be learned from these fields is that the initial engineering-centric methods are eventually supplemented with inspiration from the cognitive sciences. These supplements bring new ways to increase the autonomy of a system, to ensure its ability to deal with events that are not foreseeable at design time, and sometimes even to ensure behavior that is intuitively understandable by humans that interact with these systems. Conversely, an autonomous system (whether a robot or a vehicle) will necessarily have to interact with humans (in the case of intelligent vehicles, these include the passengers inside the vehicle, pedestrians, other vulnerable road users, and drivers of vehicles that are not automated). The vehicle must therefore also be able to understand and predict the actions of others.
Overall, such cognitively inspired approaches are now well established in robotics, both at the control level, and where interaction with human users is concerned, but are only beginning to emerge in intelligent vehicle development. The purpose of the present workshop is therefore to give a forum to researchers who either apply cognitive approaches to intelligent vehicles, or have made major contributions to robotics in this manner. The content of this workshop is thus relevant to anyone interested in one of the major under-researched areas in the field of intelligent vehicles, now ripe for exploitation.
|08:30 – 09:00||Serge Thill
Plymouth University/University of Skövde
|Introduction: Cognitive Robotics as an inspiration for Cognitive Vehicles (and vice versa)|
|09:00 - 10:00||Yiannis Demiris
Imperial College London
|Keynote: Cognitively inspired User Modelling for enhanced Human-Vehicle interaction|
|10:00 - 10:30||Coffee|
|10:30 - 11:10||Henrik Svensson
University of Skövde
|Using higher-level cognition and simulation as inspiration for intelligent vehicles.|
|11:10 - 11:50||Alex Blenkinsop
University of Sheffield
|Biologically inspired decision making for automated driving algorithms|
|11:50 - 12:30||Mauro da Lio
University of Trento
|Artificial drivers technologies for future Intelligent Vehicles and Transportation Systems|
|12:30 - 13:30||Lunch|
|13:30 - 14:10||David Windridge
|Hierarchical Perception Action learning for Cognitive Driver Assistance|
|14:10 - 14:50||Jonas Andersson
|Communicating intent for seamless interaction with automated vehicles|
|14:50 - 15:20||Coffee|
|15:20 – 16:00||Tim Tiedemann
University of Applied Sciences Hamburg
|Concept of a Cognitively-Inspired Distributed Data Processing Approach in Automotion and Its Evaluation Framework|
|16:00 - 16:40||Roberto Montanari
|Autonomous but cognitively inspired: the EU project AUTOMATE approach to vehicular automation|
|16:40 - 17:00||Conclusions|
Henrik Svensson, University of Skövde
Using higher-level cognition and simulation as inspiration for intelligent vehicles.
The paradigm of embodied cognition in cognitive science emphasizes the embodied and action-oriented nature of cognition, but some embodied cognition theories also explain the nature of higher-level cognition in terms of embodied interaction. According to the simulation hypothesis, higher-level cognition can be explained as reactivation, i.e., the brain reactivates itself as if it actually were actively controlling the body. In particular, predictive chains of simulated perceptions and actions can be reactivated internally by our nervous system and used in situations calling for representations of the future, especially alternative futures. Our and others research shows that it is possible but not trivial to develop these kinds of predictive chains of simulated perceptions and actions in simple wheeled robots. Simulations in this sense, could be used, not only in simple robots, but in more advanced types of vehicles such as cars, trucks, buses and construction machines whose environment and tasks is likely to call for more advanced mechanisms to cope with all the types of traffic situations that a human driver can. While it is not given that biological inspiration leads to the best technological solutions, this presentation describes the simulation hypothesis and outlines possible ways in which simulation theory and higher-level cognition could be used for developing mechanisms for intelligent vehicles. A brief aside also discusses the so-called anti-representationalist debate in embodied cognition and possible implications for the development of intelligent vehicles.
Mauro da Lio, University of Trento
Artificial drivers technologies for future Intelligent Vehicles and Transportation Systems
This talk will introduce the notion of artificial co-drivers. It will start with an early problem, more than 20 years ago, which was assessing the maneuverability of unstable vehicles (motorcycles) which was solved by simulating these vehicles as if they were “optimally” driven by a sort of imaginary perfect test driver. This lead to the not surprising discovery that minimum time optimal control matches the way trained race drivers actually drive, hence introducing the notion of optimal driving agents.
The talk will then shift to the problem of modeling which optimality criterion holds for ordinary drivers, introducing some theories about optimality of human control (in particular minimum jerk). It will show the use of Optimal Control to model “ideal” ordinary drivers and its application to produce “reference maneuvers” used as gold standard in Advanced Driver Assistance Systems. In particular, the application of these ideas in the PReVENT project (2007) will be presented.
The problem of modelling human driver behavior when multiple choices are possible, is then considered, reviewing the Simulation Hypothesis of Cognition and its implication for the inference of intentions of drivers (a process that in natural cognition is called mirroring – from mirror neuron theory – and which can also be considered as the “mother nature” version of model-based state estimation). The application of this mechanism in the EU InteractIVe project will be presented, showing how this mechanism naturally leads to a topographic representation of possible actions, which has direct analogies with the human motor cortex. Based on this idea, an architecture that simulates human layered control (with action selection) will be presented and, it will be shown how the mirroring process can be implemented with a particular type of action-selection mechanism. Hence the same sensorimotor system can be used to engineer an agent that drives autonomously or that understands the intentions of the human driver and cooperates in the driving tasks. In the conclusions, the potential impacts of the reviewed technologies will be examined.
David Windridge, Middlesex University
Hierarchical Perception Action learning for Cognitive Driver Assistance
Perception-Action (P-A) learning is an approach to cognitive system building that seeks to reduce the complexity associated with conventional environment-representation/action-planning approaches. Instead, actions are directly mapped onto the perceptual transitions that they bring about, eliminating the need for intermediate representation and significantly reducing training requirements.
Subsumptive P–A models capture the hierarchical nature of the task structuring implicit in human agents and further assume a parallel hierarchical structuring within the agent's perceptual domain. In the context of an intelligent driver assistance system, adopting this model enables us to characterize intentions at each level of the P–A hierarchy in terms of perceptual descriptors correlated with driver behavior. A key problem in the context is reconciling high-level protocols (i.e., Highway Code rules) with low-level perceptual features; we here set out a general approach to the integration of abstract symbolic manipulation into P-A learning.
Jonas Andersson, RISE Viktoria
Communicating intent for seamless interaction with automated vehicles
Mutual understanding and trust between automated vehicles and their occupants, as well as between automated vehicles and other road users in their vicinity, such as pedestrians and bicyclists, are two cornerstones for seamless and safe interactions in our traffic. This presentation highlights the importance of clear communication among all involved parties, both inside and outside vehicles, for creating such interactions. In particular, it presents a series of studies conducted by RISE Viktoria and partners in Sweden, where it is hypothesized that communicating own intent makes it easier for automated vehicles and humans to predict and understand each other’s behavior and to act accordingly. This is explored in different contexts using both interior and exterior vehicle interfaces for intent communication.
Roberto Montanari, Re:Lab s.r.l.
Autonomous but cognitively inspired: the EU project AUTOMATE approach to vehicular automation
In autonomous systems, a human-like understanding of environment, as well as mutual comprehension between drivers and machine, is expected to significantly improve a seamless and effective transition in vehicle controls, rightly allocating the driving task to the agents (no matter if human or machine) most suitable and reliable at each moment. The steps to reach this ambitious goal are in the scope of AUTOMATE project, and mostly in the part dedicated to the Human Machine Interface design, one of the core of this EU funded project within the umbrella of Horizon 2020. This presentation will give an overview of the project's current work, focusing the attention on the paradigms adopted in machine learning techniques framework, in human factors and in HMI design.