Skip to main content

Researchers develop new AI to enable autonomous vehicles to adapt to challenging weather conditions


A new AI uses self-supervised learning to tackle the most challenging aspects of robust positioning in autonomous driving caused by adverse weather conditions.

Researchers in the Department of Computer Science, in collaboration with colleagues from Bogazici University, Turkey, have developed a novel AI that helps AVs to achieve safer and more reliable navigation capability, especially under adverse conditions and GPS-denied driving scenarios.

The researchers start by questioning 'why autonomous vehicles (AVs) are still limited to relatively small-scale trials, although several leading AV companies predicted that AVs would be on the roads by 2020'. They argue that a major limitation comes from adverse weather. Almalioglu and colleagues propose a novel self-supervised deep learning architecture for ego-motion estimation, a crucial component of an autonomous driving algorithmic stack. Researchers use a geometry-aware learning technique that fuses visual, lidar and radar information, such that the benefits of each can be used under different weather conditions.

Yasin Almalioglu, who completed the research as part of his DPhil in the Department of Computer Science, University of Oxford said:

'Our AI acts as complementary software to improve the safety and reliability of AVs under adverse weather conditions, providing precise positioning under challenging conditions. To understand the importance of precise positioning, think about An AV that might detect itself in the wrong lane before a turn, or might stop too late at an intersection because of imprecise position. We propose an alternative solution to achieve such precision for safe and reliable AVS in challenging environments such as rainy, foggy or snowy.'

Professor Andrew Markham, who co-supervised the research, said:

'Estimating the precise location of AVs is a critical milestone to achieving reliable autonomous driving under challenging conditions. This study effectively exploits the complementary aspects of different sensors to help AVs navigate in difficult daily scenarios.'

Professor Niki Trigoni, who co-supervised the study, said:

'The precise positioning capability provides a basis for numerous core functionalities of AVs such as motion planning, prediction, situational awareness, and collision avoidance. This study provides an exciting complementary solution for the AV software stack to achieve this capability.'

The scientists anticipate that this work will bring AVs one step closer to safe and smooth all-weather autonomous driving, reserving a broad impact on AV applications in society.

The full paper, Deep learning-based robust positioning for all-weather autonomous driving, is published in Nature Machine Intelligence.