Wayve

Wayve is a leading developer of artificial intelligence (AI), pioneering Embodied AI technology for the automotive industry. The firm’s next generation approach enables their Embodied AI to learn to drive using the broadest range of data from vehicle cameras and sensors to more general-purpose knowledge from the internet.

The sensor and hardware-agnostic AI software can quickly and safely adapt its driving intelligence to new, unseen environments without needing expensive sensors and high-definition maps.

Co-founder and CEO, Alex Kendall, explains how Embodied AI is going to transform not only transportation but the full gamut of tools and tasks in the physical world.

What was your inspiration for founding Wayve?

I'm inspired by what Embodied AI is going to do for the world. I think it's going to be the most incredible transformation that we go through in our lifetimes – the ability to bring AI into the physical world.

If you think about the history of the tools and machines that we've developed as a human race, Embodied AI gives us the ability for them to interact with us in the physical world in a way that we can trust, in a way that makes our lives more safe, sustainable, and enjoyable.

Why did you first focus on self-driving cars?

Self-driving cars will be the first example of Embodied AI for two main reasons.

Firstly, there is an overwhelming need. Road accidents claim 1.3 million lives a year globally and over 99.6% of those accidents are due to human error. And, apart from the human tragedy, the economic cost of these accidents to the United Kingdom and the United States is about 1.5% of each overall GDP which is enormous – we’re talking about billions.

So, we have a real need and a real opportunity to eliminate these deaths and make transportation safer, cheaper, more accessible, and more sustainable.

The second factor – and this is a real ‘why now’ moment – is that the leading automotive manufacturers have all invested in cars that are now at consumer production volumes that have the onboard sensors and computational power to be able to run a system like ours.

Can you explain how your system works and how it differs from other approaches?

The automotive market has mostly focused on autonomous driving systems that are told how to drive. They follow a fixed map and have a rule set that is hand coded with how they should behave.

These systems operate at extraordinary expense and without the general-purpose nature needed to be able to deal with all the unusual events that can occur on the road. Driving in the physical world is open-ended. It’s not rule driven. There are so many changing factors interacting in real time.

Our starting point was that it would be better to build a data-driven AI system that can learn about and understand the world in a more general way. So, we have created a single neural network and taught it how to drive and be able to make sense of all that complexity and diversity.

How do you train the AI?

We train our system from as diverse a variety of data sources as possible – from autonomous vehicles that we operate, dash-camera footage from our partners’ vehicles, and even robots that aren't vehicles. There are tens of millions of vehicles driving around in the world right now that have sensors on them that we can use to learn and train the AI.

It also learns directly from the internet. Our AI literally reads the PDF document that the UK Government published on the road code, it reads text to learn about concepts like cars and roads and risk and human values, but it also reads a broader range of information on science, mathematics, literature, ethics, etc. So, it can understand concepts that go far beyond its training data to have a general-purpose intelligence.

Apart from safety benefits, can you give us a sense of how our lives will be improved through a more autonomous transport system?

Firstly, fully autonomous vehicles will give us a driving experience that is delightful – a true chauffeur experience. There'll be all sorts of amazing infotainment experiences built out in these cars. Maybe you want to sleep. Maybe you want to read a book or play with your kids. Think also about car design and interior layouts that have always been based on there being a driver.

Think about car parking. Cars are only used about 3% of the time. So right now, they are stationary 97% of the time. On our streets. Outside our houses. In car parks.

In London, currently, car parks take up an area 10 times the size of Hyde Park and all this space currently occupied by unused cars can be freed up through autonomy. Autonomous cars can park themselves and charge out of the way in areas of low-value real estate. And when you need your car or any car you can use autonomous ride-hailing systems that summon the vehicle to you.

Take public transport as well. Public transport can’t be as frequent as maybe people would like because the cost is high. With autonomous transport, you end up with a system that is frequent, efficient, and personalized.

I think it's important to have a heterogeneous transportation system from walking to biking to micromobility to trains to cars, and there are various ways that AI can help optimize all of those.

We have an opportunity to create much healthier cities. I honestly believe that autonomy in transport is going to be the biggest change that we've seen since the invention of the car 140 years ago or so.

What are further uses for Wayve’s technology outside of autonomous vehicles?

We are totally focused on self-driving vehicles – the data is there, the AI is there and the vehicles are ready and that’s what will allow us to bring autonomous driving to cities everywhere.

But because our AI is of a general type rather than a particularly focused system means that it can learn from other data and be applied in other situations. So, we can imagine applying this AI to other kinds of robotics, from manufacturing to domestic robots, to be able to create a true age of autonomy.

I think Embodied AI will provide us with tools that we can choose to use or not. Tools that we can delegate tasks to so we can free up our time for what matters most to us — removing the need for mundane or dangerous work and freeing up time for more human connection, creativity, and value.

I dream of a world where we have intelligent machines and tools that we can trust to be able to do the work that we don't want to do.

It’s interesting to look at how AI has developed. It's gone from 10 years ago, when the focus on computer vision drove all the big breakthroughs, to the last five years, when it has been mainly about language. I think robotics is going to be the next one because it's bringing that next level of scale of data, petabytes of video data. And it's bringing the really challenging problems that will truly force us to build new and better AI-enabled systems.

What these machines look like and how they function is still up for grabs. Will they be very task-specific or more general? Will they look humanoid or totally different from our expectations? But we will have intelligent machines that can assist us and augment all these aspects of our lives.

We still need to build the robots, and we need to get the AI adapted, and that needs data. But it will come. It's growing. There's a market for that. And I think the AI we build in driving will be more general purpose and intelligent and allow us to enter these new spaces.