By: Paul Myles, TU Automotive
Connectivity is being heralded as the only way forward for the autonomous vehicle, yet Paul Myles found out that a group of Oxford scientists are proving their car can navigate without the Internet’s help
It’s probably of little surprise that the scientists behind the Mars Rover approached designing an autonomous vehicle without the imperative of having it communicate with the Internet.
And so boffins from the ‘dreaming spires’ of Oxford in the UK have devised a system that can allow any vehicle to navigate its way through the tiny, higgledy-piggledy streets and lanes of the ancient seat of learning completely free from the shackles of connectivity.
While some sectors of the auto and software industries might throw their hands up in horror at the prospect of a vehicle whose intelligence comes from within itself, severing the lucrative data stream to commercially interested third parties, the system could, in practice, accelerate the early adoption of autonomy in cars.
TU-Automotive went to find out more from Dr Graeme Smith, chief executive of Oxbotica, the spin-out of Oxford University’s Mobile Robotics Group responsible to commercialising the group’s advanced work in autonomous vehicles.
Smith took up story of how this group devised unique mathematical algorithms which allow three dimensional map building using simple two dimensional sensors.
He said: ‘Oxford Mobile Robotics Group has been pioneering innovative research in the area of mobile autonomy for a number of years now. They have conducted more than 100 man years of research into this area and have developed a huge portfolio of intellectual property (IP) around the autonomous vehicle.
‘The IP is very practical, for example downstairs in the garage we have the UK’s only licenced driverless car. There’s only one.
‘Based on a Nissan Leaf, clearly the robotics group didn’t build the car but what they did do is develop the autonomous control system and integrate the sensors that allows it to drive autonomously.
‘One of Oxbotica’s roles is to commercialise that IP and take it to market,’ Smith explained. ‘We have negotiated exclusive worldwide rights for most the group’s intellectual property and our remit is to find channels to market with this.
‘We have only been in business since September 2014 and we see ourselves not as a product company but as a licensing and IP company, working with customers to help them to integrate this IP into their products. This means we work with them on engineering support, technology transfer and on licensing.
‘We are all about supporting customers and helping them to market whether it be in a car, a warehouse, or round a closed test track in any form of autonomy. What we are also finding out that there is a huge spin-out of opportunities for a lot of this technology. For example, when NASA went to the moon we are still deriving the benefits of that investment into technology all these years later.
‘There is a lot of first class research taking place around the autonomous car and there are spin-out technologies that are very attractive to other companies. One example is our ability to quickly survey a city map in 3D is of extreme interest to surveying companies, whether it civil engineering for street surveys, road surveys or an internal survey from being able to do a 3D survey very, very quickly.
‘It has been possible to do 3D surveys using static equipment but it takes a long time and those companies that have mobile survey platforms typically cost about £300,000. We are probably just 5% of that cost. We may not perform at exactly the same level of resolution as one of those high-end systems but for 5-10% of the price if you can get somewhere near achieving say 80-90% of the functionality then that enables a lot more business models that, perhaps, didn’t seem worthwhile at the higher cost point.
‘We have had a lot of interest from companies that thought this would not be possible for them to afford this technology but now they are developing their own ideas around products that benefit from cheap, quick surveys.’
Smith said the ‘magic’ lies within the mathematical algorithms which allow simple cheaply sourced sensors to build up a sophisticated three dimensional map to rival that of others more costly systems currently being employed by big organisations.
He explained: ‘We are using commercially available off-the-shelf sensors and a lot of our IP is around how you synchronise those sensors together and keep them synchronised. We use a technique called visual odometry that helps you understand how you move in space, a bit like a pair of eyes, and then we use 2D lasers, rather than expensive 3D ones, that helps us map what is either side of us. As we move through the space, we are able to create the 3D model.
‘Oxbotica’s approach has been radically different from other approaches in the market that are a lot more complex but a lot more expensive as well. The core of our offering is the algorithms, the software and the thought behind them.’
And, beyond its Oxford mule, Oxbitica is employing this technology in trials of autonomous ‘pods’ in the UK’s Midlands.
Smith said: ‘We are working as part of the government’s driverless car challenge in Milton Keynes and Coventry that is called the UK Autodrive. As part of that we are working on the control systems that will be going into the 40 pods. These are made by a company in Coventry called the RDM Group and we help them integrate the sensors and then the entire control system which has been designed to cope with the same sort of environment that the pods will work in.’
These pods use 3D maps of their environment recorded throughout the seasonal changes they will have to handle.
He said: ‘The way the pods will work is that we have mapped the cycle ways where these pods will work multiple times in rain, in snow in sun and in summer and winter. From this we have created 3D models of the pathways in all these different conditions and the pods will be able to work within this map using their sensors to localise themselves within the map to know where they are to within a few centimetres. They then navigate along the routes that we have pre-mapped.
‘These pods will also have the ability to continually map as they go so we can upgrade the map if we wanted to. For other applications it’s possible that, if you are thinking ahead to 20-years’ time when a car is autonomous, one school of thought is that the car could come equipped with 3D maps of the world. Just in the same way you can buy CDs of maps at the moment, you’ll be able to buy 3D map CDs.’
Smith said an alternative approach would be a process of ‘teaching’ the car to recognise an individual motorist’s often used journey routes.
‘It could be that your car is creating a map itself as it drives,’ he said. ‘Possibly, when you buy the car in the first place the map is empty but when you drive a route to work it starts to learn about that route. Perhaps on day two or three a little light on the dashboard will tell you ‘I remember this bit and am OK to drive this bit if you like?’ In this way, over a period of time, you effectively create your own database.’
One of the key strengths of the system is the way the technology can use 3D point clouds to distinguish between the fixed obstacle and that which is temporary.
He said: ‘When we are navigating we want to make sure we are localising from things in the environment that are static. We have a lot of software, for example, that will start to remove things from the scans such as parked cars, pedestrians and anything that we determine are transient. If we scan the same route multiple times, we start to learn about which things are there all the time and which are not. So, again, by being able to over-lay one scan on the next we are able to build the database and eventually get back to something that is very static, which is the best thing to navigate from.’
This, effectively, liberates the vehicle from the need to be hooked up to the Internet, a strength that makes autonomy possible in areas of poor GPS coverage, whether that be in remote rural areas or in city skyscraper canyons.
Smith said: ‘The approach that we have is completely GPS-free and we don’t use any sort of infrastructure beacons to locate the vehicles. Those things may come along but everything that we have done has assumed we are infrastructure free that means it is equally applicable to a car or a mining robot or a one inside a shopping centre or even a train. In this way the technology is completely cross-platform.’
However, Smith had some good news for those still commercial interests whose business models are dependent on an autonomous vehicle being connected to the Internet.
He added: ‘This doesn’t rule out cars communicating with each other, sharing information and databases, and certainly doesn’t rule out some centralised infrastructure to download local sections of the map.
‘We don’t know right know, so what we are doing is working on a level below that and we are able to build our own maps in real time and to build them ahead of time and navigate with them.’
Smith believes the autonomous vehicle is unlikely to burst suddenly onto the scene in the older cities of Europe but rather embark a more gradual process of integration with existing transport infrastructure.
He said: ‘We think we will see autonomy adopted in stages and we already know that new cities being built in the Far East they are thinking about autonomous transportation and may well design a city around it. In that environment of a purpose built city with a purpose built autonomous transport system, we can see elements coming in much more quickly. Certainly, closed environments like warehouses there are already autonomous solutions and we can see this creeping into many different dimensions.
‘However, there will always be challenges in introducing new technology in a mixed environment – introducing autonomous driving into an environment that wasn’t designed for it with other forms of transport might be more problematic. I think it might take some time and until the industry is able achieve a critical mass to make this successful.
‘To start with, we may have an autonomous lane on a motorway or even a dedicated autonomous motorway.
‘It’s easy to predict how one autonomous car would interact with another autonomous car but it’s more complex to think how it would react to a human driver or cyclist. It’s this interaction that would be the main reason for a slow implementation of the technology.’
This article first appeared in TU Automotive in July 2015 with the title Smart car for the dumb city.