TASCC: The Cooperative Car
Lead Research Organisation:
University of Warwick
Department Name: Computer Science
Abstract
We are at the start of, arguably, the most significant transition in motoring for a century as the complex tasks involved in driving become increasingly performed by machine. Individual drivers and their cars will form part of wider and smarter urban transport infrastructure, and the cars of the future will be intelligent and cooperative.
The opportunities to deliver better safety, traffic efficiency, and more productive and pleasant journeys are enormous, but a revolution on this scale faces great challenges for science and society.
Almost imperceptibly to the driver, modern vehicles are equipped with hundreds of micro-computers and sensors, including cameras, radar, GPS, and telemetry measuring everything from speed, braking, and steering to environmental conditions. Many vehicles have wireless communications (from 2018, new EU cars will have data communication for automated emergency calls) enabling data to be uploaded in real-time to the cloud to be later analysed and used. Current vehicle features operate relatively independently, however such data gives the potential for a vehicle to learn about its driver and environment, and paves the way for integrated intelligent features and eventually for autonomous cars. Despite significant progress, there are many unsolved challenges, not least related to how such cars will be accepted by the public. So far, autonomous vehicles have been confined to small geographic areas, for example Google's Self-Driving Car relies on detailed data prepared beforehand by human and computer analysis, and is unable to fully cope with adverse weather, road works and other real-world aspects of driving. There has been thus far little research on: how autonomous vehicles will fit in with today's manually driven cars; how drivers and occupants will interact with them; and how they will run safely in our towns, with pedestrians and cyclists.
Accelerating the transition to autonomous vehicles, this project will tackle scientific challenges whose solutions will deliver some of the convenience, safety and efficiency benefits of future autonomous cars in mainstream vehicles, and will lay the foundation for fully autonomous vehicles. Jaguar Land Rover has a vision of a self-learning car (SLC) that will minimise driver distractions, enhance safety, and deliver a personalised driving experience. In this project, we will apply advanced research techniques in machine learning and the processing and mining of large data streams to make the SLC a reality. For example, we will use telemetry and information about the occupants, such as their cognitive load, to personalise the driving experience, predict the destination, adaptively configure safety systems, advise on congestion avoidance and parking opportunities.
In the near future vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2X) communication will be a reality: cars will know about other cars on the road and be able to exchange information with them. Cars will become cooperative: with each other and with urban environments. When combined with existing sensors, vehicles will be able to share information on road, traffic and parking conditions. This project will develop software algorithms, applying experimental methods from behavioural sciences and processing information from connected cars to understand driver habits, and develop strategies to encourage behaviour modifications: for example to design adaptive pricing to reduce parking and congestion. We will also investigate how best human drivers and autonomous cars can interact, for example when taking or handing-over control, or when interacting and negotiating with other road users.
In order to deliver safe and efficient autonomous and semi-autonomous cars of the future, we will develop intelligent driver systems, and cooperation and behaviour modelling techniques that learn about drivers, enabling vehicles to cooperate with each other and with urban transport infrastructures.
The opportunities to deliver better safety, traffic efficiency, and more productive and pleasant journeys are enormous, but a revolution on this scale faces great challenges for science and society.
Almost imperceptibly to the driver, modern vehicles are equipped with hundreds of micro-computers and sensors, including cameras, radar, GPS, and telemetry measuring everything from speed, braking, and steering to environmental conditions. Many vehicles have wireless communications (from 2018, new EU cars will have data communication for automated emergency calls) enabling data to be uploaded in real-time to the cloud to be later analysed and used. Current vehicle features operate relatively independently, however such data gives the potential for a vehicle to learn about its driver and environment, and paves the way for integrated intelligent features and eventually for autonomous cars. Despite significant progress, there are many unsolved challenges, not least related to how such cars will be accepted by the public. So far, autonomous vehicles have been confined to small geographic areas, for example Google's Self-Driving Car relies on detailed data prepared beforehand by human and computer analysis, and is unable to fully cope with adverse weather, road works and other real-world aspects of driving. There has been thus far little research on: how autonomous vehicles will fit in with today's manually driven cars; how drivers and occupants will interact with them; and how they will run safely in our towns, with pedestrians and cyclists.
Accelerating the transition to autonomous vehicles, this project will tackle scientific challenges whose solutions will deliver some of the convenience, safety and efficiency benefits of future autonomous cars in mainstream vehicles, and will lay the foundation for fully autonomous vehicles. Jaguar Land Rover has a vision of a self-learning car (SLC) that will minimise driver distractions, enhance safety, and deliver a personalised driving experience. In this project, we will apply advanced research techniques in machine learning and the processing and mining of large data streams to make the SLC a reality. For example, we will use telemetry and information about the occupants, such as their cognitive load, to personalise the driving experience, predict the destination, adaptively configure safety systems, advise on congestion avoidance and parking opportunities.
In the near future vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2X) communication will be a reality: cars will know about other cars on the road and be able to exchange information with them. Cars will become cooperative: with each other and with urban environments. When combined with existing sensors, vehicles will be able to share information on road, traffic and parking conditions. This project will develop software algorithms, applying experimental methods from behavioural sciences and processing information from connected cars to understand driver habits, and develop strategies to encourage behaviour modifications: for example to design adaptive pricing to reduce parking and congestion. We will also investigate how best human drivers and autonomous cars can interact, for example when taking or handing-over control, or when interacting and negotiating with other road users.
In order to deliver safe and efficient autonomous and semi-autonomous cars of the future, we will develop intelligent driver systems, and cooperation and behaviour modelling techniques that learn about drivers, enabling vehicles to cooperate with each other and with urban transport infrastructures.
Planned Impact
This project will deliver better safety, traffic efficiency, and more productive and pleasant vehicle journeys. Our aim is to develop the algorithms and technology required to: (i) mine on-board, physiological and occupant device data to provide intelligent and autonomous vehicle features, (ii) share and mine off-board data from other vehicles and infrastructure to provide cooperative vehicle features, and (iii) develop cooperation, virtual bargaining and profiling techniques to enable creation of incentives/sanctions and marketplaces to influence driver behaviour and inform OEM analytics.
The resulting algorithms and technology will be directly applicable to Jaguar Land Rover (JLR), the automotive industry, and to other business contexts and the wider scientific community. In the short term, the project addresses challenges raised by the desire for intelligent features such as the Self-Learning Car (SLC). In the medium term the project will feed into vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2X) features, including advanced collaborative intelligent applications (e.g., smart navigation), and will inform usage pattern analysis and fault prediction. Longer term our behavioural analysis and understanding of the human factors involved in influencing behaviour will impact on developing smart-city technologies such as adaptive road/parking pricing for management of congestion, noise, and emissions and OEM strategies and analytics, including personalised pricing strategies for warranty/insurance services.
Given widespread vehicle use and the importance of issues such as congestion, noise and emissions, this research has significant strategic and societal importance that goes beyond the immediate impacts for JLR in SLC and V2X applications. Understanding human behaviour is essential for the sucessful development and acceptance of autonomous vehicles and systems, and the project will have impact in this respect. The project will have direct impact in terms of safety and efficiency, and will have broader economic and societal benefits, e.g., increased safety can reduce the emotional impact and monetary cost of road accidents. By developing the algorithms and technology needed to make driving easier and less stressful, the project will contribute to well-being by improving the general health and state of vehicle occupants.
Although the project is directed at the automotive sector, our research will impact the wider scientific community, particularly in the areas of (i) machine learning and data analysis, (ii) psychology of cognitive load, distraction and acceptance of autonomous systems, (iii) sketch and scalable computing techniques, and (iv) behavioural science.
Specifically, the project will result in:
1. New data capture, streaming, and feature extraction methods for vehicle, city-level and human data, that are of general applicability.
2. A machine learning toolkit for such data that will shorten development cycles of customer-facing features and improve integration (of immediate benefit to SLC and V2X applications).
3. Persona profiling and pattern-of-life prediction models that offer a richer view of driver behaviour than is currently available, and provide longer term insight into how humans interact with autonomous vehicle features. This will have direct impact on the field of behavioural science, contributing knowledge about human behaviour, how it can be nudged and influenced, and how best to enable humans and machines to cooperate safely and efficiently.
4. Determining factors that influence driver behaviour to inform development of effective driver modification strategies. This will support advanced smart-city and V2X features that impact on a range of business areas and city-level problems.
There is great potential for IP generation, and the appropriate exploitation of IP will be considered as part of the project management.
The resulting algorithms and technology will be directly applicable to Jaguar Land Rover (JLR), the automotive industry, and to other business contexts and the wider scientific community. In the short term, the project addresses challenges raised by the desire for intelligent features such as the Self-Learning Car (SLC). In the medium term the project will feed into vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2X) features, including advanced collaborative intelligent applications (e.g., smart navigation), and will inform usage pattern analysis and fault prediction. Longer term our behavioural analysis and understanding of the human factors involved in influencing behaviour will impact on developing smart-city technologies such as adaptive road/parking pricing for management of congestion, noise, and emissions and OEM strategies and analytics, including personalised pricing strategies for warranty/insurance services.
Given widespread vehicle use and the importance of issues such as congestion, noise and emissions, this research has significant strategic and societal importance that goes beyond the immediate impacts for JLR in SLC and V2X applications. Understanding human behaviour is essential for the sucessful development and acceptance of autonomous vehicles and systems, and the project will have impact in this respect. The project will have direct impact in terms of safety and efficiency, and will have broader economic and societal benefits, e.g., increased safety can reduce the emotional impact and monetary cost of road accidents. By developing the algorithms and technology needed to make driving easier and less stressful, the project will contribute to well-being by improving the general health and state of vehicle occupants.
Although the project is directed at the automotive sector, our research will impact the wider scientific community, particularly in the areas of (i) machine learning and data analysis, (ii) psychology of cognitive load, distraction and acceptance of autonomous systems, (iii) sketch and scalable computing techniques, and (iv) behavioural science.
Specifically, the project will result in:
1. New data capture, streaming, and feature extraction methods for vehicle, city-level and human data, that are of general applicability.
2. A machine learning toolkit for such data that will shorten development cycles of customer-facing features and improve integration (of immediate benefit to SLC and V2X applications).
3. Persona profiling and pattern-of-life prediction models that offer a richer view of driver behaviour than is currently available, and provide longer term insight into how humans interact with autonomous vehicle features. This will have direct impact on the field of behavioural science, contributing knowledge about human behaviour, how it can be nudged and influenced, and how best to enable humans and machines to cooperate safely and efficiently.
4. Determining factors that influence driver behaviour to inform development of effective driver modification strategies. This will support advanced smart-city and V2X features that impact on a range of business areas and city-level problems.
There is great potential for IP generation, and the appropriate exploitation of IP will be considered as part of the project management.
Publications
Bradbury M
(2020)
Privacy Challenges With Protecting Live Vehicular Location Context
in IEEE Access
Chater N
(2018)
Negotiating the Traffic: Can Cognitive Science Help Make Autonomous Vehicles a Reality?
in Trends in cognitive sciences
Cormode G
(2017)
Data sketching
in Communications of the ACM
H Mckay
(2019)
Online Transfer Learning for Concept Drifting Data Streams
J Misyak
(2019)
Joint action in driving: A virtual bargaining approach
McKay H
(2020)
Bi-directional online transfer learning: a framework
in Annals of Telecommunications
P. Taylor
(2022)
Feature selection for supervised learning and compression
in Applied Artificial Intelligence
Ritchie O
(2019)
How should autonomous vehicles overtake other drivers?
in Transportation Research Part F: Traffic Psychology and Behaviour
| Description | There are five main areas in which contributions have been made, as follows. 1. We developed methods for feature selection which not only consider the performance of the resulting features when used to build a machine learning model, but also consider compressibility. This enables system designers to balance the tradeoff between the data size required (which often needs to be stored for archive or to be communicated) and the performance of the resulting models. 2. We developed a novel compression technique for vehicle data. This is currently subject to a patent application by the industry partner. Publication of the technique is pending and awaiting the patent application. 3. We created a method for predicting a vehicle's activity based primarily on telemetry data, and subsequently applied this to improve the performance of destination prediction algorithms. 4. We investigated the psychological impact, in terms of satisfaction, of the traffic context of an autonomous vehicle and the way that information is presented to a driver/passenger. Our results show that not only is the traffic context important, but also that the way information is presented has a significant impact of user satisfaction. 5. We developed a method for enabling bi-directional transfer learning, meaning that models learnt in one application are transferred to other applications. Previous approaches have be one-directional and have assumed that source applications are offline. As part of this work, we also developed scalable methods for model selection and assessing whether to transfer models (to reduce computation and communication overheads). |
| Exploitation Route | Our results on feature selection, compression, and transfer learning are likely to be useful in the future development of applied machine learning approaches for real-world applications. The work on the psychological aspects of satisfaction of autonomous vehicles is expected to be useful to vehicle manufacturers in developing the way information is presented to users of self-driving vehicles. Our work on activity and destination prediction is also likely to be useful to vehicle manufacturers in developing future intelligent vehicle functionality. |
| Sectors | Digital/Communication/Information Technologies (including Software),Transport,Other |
| Description | Given the nature of the project the majority of non-academic impact has been directed towards the industry partner. The primary areas of impact are as follows. 1. Our work on user satisfaction in the context of autonomous vehicles and intelligent systems is informing human factors work at the industry partner. 2. Our compression-aware feature selection methods have been used by the industry partner. 3. The novel compression methods developed during the project is in the process of being patented by the industry partner. |
| First Year Of Impact | 2021 |
| Sector | Manufacturing, including Industrial Biotechology,Transport |
| Impact Types | Economic |
| Description | Nathan Griffiths' membership of the Royal Society Machine Learning Working Group |
| Geographic Reach | National |
| Policy Influence Type | Participation in a guidance/advisory committee |
| URL | https://royalsociety.org/topics-policy/projects/machine-learning/ |
| Title | LED: Location Exraction Dataset |
| Description | The Location Extraction Dataset (LED) is a high-resolution vehicular dataset for investigating point of interest extraction |
| Type Of Material | Database/Collection of data |
| Year Produced | 2019 |
| Provided To Others? | Yes |
| Impact | This dataset has supported the development of a novel location extraction technique, and is being used in the development of more advanced techniques. The LED is the dataset on which the following publication is based: James Van Hinsbergh, Nathan Griffiths, Phillip Taylor, Alasdair Thomason, Zhou Xu, and Alex Mouzakitis. 2018. Vehicle Point of Interest Detection Using In-Car Data. In 2nd ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery (GeoAI'18), November 6, 2018, Seattle, WA, USA. ACM, New York, NY, USA, 4 pages. https://doi.org/10.1145/3281548.3281549 |
| URL | https://www.dcs.warwick.ac.uk/led/ |