Autonomous cars. How do they work? How much do we trust them? And what role might they have in our future transport systems?
Machine learning is a key part of driverless car technology; this is a way of programming computers so that, instead of giving a computer a set of detailed instructions to follow, you can program it to be adaptive, and learn from data.
Sabine Hauert, a Lecturer in Robotics from the University of Bristol and working group member for the Royal Society’s machine learning project, chaired the discussion. Joining Sabine were; Paul Zanelli, Chief Technical Officer at Transport Systems Catapult; Maja Pantic, machine vision researcher; and Ingmar Posner, machine learning and robotics researcher.
You can see more from our panellists in our video.
So how do driverless cars work?
Driverless cars use sensors to collect data from their environment so that they can understand their surroundings, and react accordingly. These onboard sensors include: lidar sensors that bounce pulses of light off nearby objects to help to detect the edges of roads and identify lane markings; and radar sensors which monitor the position of nearby vehicles. There are also cameras on the vehicles.
Machine learning helps process the data from all these sensors, so that the car can make sense of what is going on around it. Learning from this data helps the car to answer three key questions: where am I; what is around me; and how should I act?
Although many people might envisage a future where driverless vehicles take the place of personally-owned cars – like the ones we use today – there are alternative potential uses of these vehicles. For example, there could be specifically designed ‘pods’, which would look different to our current vehicles, and which could travel on pavements as well as roads. These different approaches are associated with different timescales.
Looking ahead, the panel discussed how driverless vehicles could change both how we use transport and how we use urban environments. With pollution in cities at non-sustainable levels, and 20 per cent of the city used for parking, Paul Zanelli argued that driverless cars could help to improve our environment and save space in cities, creating efficiency savings – in terms of transport and health – as a result.
Questions from the audience
In considering the future of this technology, the audience had questions about how driverless cars can follow social values, and their security. Ingmar Posner discussed how driverless vehicles could be trained to follow an ethical framework, partly by learning from decisions the driver made. Maja Pantic noted that behavioural research would be important when teaching driverless cars to respond to pedestrians and other vehicles on the road, as human movement can be difficult to predict.
When it came to security, and the resilience of driverless cars to hacking, the panel recognised that the UK is a world leader in ‘physical’ and ‘logical’ security so we are well placed to deal with cybersecurity threats.
The UK Government is currently consulting on whether to establish a driverless car test facility to support the development of autonomous vehicles. Such testing facilities could be used to explore some of these issues, by looking at how driverless cars fare in ‘real world’ environments.
The panel agreed that experts in the field, policy-makers and the public would need to work together so that these vehicles could be designed to act in a socially-acceptable way.
The Royal Society is currently carrying out a project on machine learning, which is considering the potential of this technology, and the challenges that come with it. You can read more about machine learning in an interview with Marcus du Sautoy.