Home • News • Case Studies • People • Publications
« Overview
source: http://archive.darpa.mil/grandchallenge/images/photos/high_res/IMG_3419_B.JPG

Long Term Infrastructure Perspective. Autonomous Driving.

Autonomous vehicles are playing an increasingly important role in our daily lives, and we require more reliable, more flexible vehicles that can carry out a variety of tasks in our complex transport infrastructure. A major challenge in the development of such autonomous systems is managing unpredictable interaction with the environment, which is ubiquitous in an urban context where vehicles are exposed to changing traffic conditions, obstacles and hazards. This case study is inspired by the DARPA Urban Challenge, a competition for autonomous cars to navigate safely and effectively in an urban setting.

Safe and Sound. Formal Methods.

Reliability and safety are of particular importance in the design of controllers for autonomous vehicles operating in situations where an accident could be fatal. Leveraging the power of formal methods in developing the high-level control capabilities allows for the provision of guaranteed performance to clients in hazardous traffic scenarios. Moreover, trade-offs between several priorities, such as minimising energy consumption, trip time, and probability of accidents can be accurately quantified. The figure on the left shows Pareto points representing optimal trade-offs between such priorities.

Managing Uncertainty. Stochastic Games.

A vehicle in urban traffic only has control over its own actions, and cannot influence other cars, pedestrians, traffic lights, or other potentially hazardous factors. While sometimes the likelihood of certain events can be predicted, a controller has to be able to appropriately handle all emergent situations. It is acting as if playing a game it wants to win against the environment, where the objective is to obtain an optimised trade-off between objectives – for example, minimise the probability of hitting an obstacle and maximise road quality. Stochastic games provide a means to model an autonomous vehicle in an adverse environment, and solve the resulting game by providing strategies that attain the promised trade-offs. A strategy is just a controller that can then be implemented in the autonomous vehicle to make the required steering decisions. We have modelled the scenario of a car driving through various English villages and show that the controllers obtained from a multi-objective extension of PRISM-games, a tool developed for the analysis of competitive systems, react correctly in the required traffic situations. In an effort to cope with larger map sizes, we are developing methods to efficiently find safe controllers by putting together controllers for smaller, local, environments.


 

Sort by: date, type, title

11 publications:

2016

2015

2014

2013

Sort by: date, type, title

« Overview