Driving towards AV2.0

Part of what has been termed AV2.0, Wayve is one of a number of start-ups consciously moving away from the robotics approach that has to date dominated autonomous vehicles. While 3D maps and separate modules are relied upon for sensing and planning, AV2.0 focusses instead on AI and training a 'deep learning' model to ‘drive’ entirely from scratch.

Indeed, less than a year after successfully demonstrating that such a model can operate in multiple UK cities, Wayve has now shown that the same AI driver can be taught to drive multiple vehicles – in this instance, a passenger car and a delivery van.

Trained using a combination of reinforcement (trial and error) and imitation (copying human behaviour) learning, it took thousands of hours to train the model to drive the passenger car. Amazingly, it then only took an additional 80 hours of data to drive the van.

By not having to keep complicated maps up to date or maintain hand-crafted software systems, it is hoped that this approach will allow Wayve to scale up faster than the current market leaders who, whilst they have brought autonomous vehicles to numerous city streets around the world, still rely on the aforementioned robotics model which has proved both expensive and prohibitive to rapid growth.

Clearly, the road to full autonomy (Levels 4 and 5) remains a long one – and one which is likely to face legal and regulatory challenges along the way. However, it is exciting to see the huge strides that are being made within the UK towards making this a reality, ultimately seeking to create an industry that has the potential to not only generate significant employment opportunities, but also growth for the UK economy. For now, watch this space.