Artificial Intelligence (AI) in automotive is well established in self-driving cars. But how far can AI stretch beyond advanced driver-assistance systems (ADAS) and play a much wider role in enhancing the software development process and customer experience?
AI is evolving at such a pace that the latest research promises to deliver far reaching impact that will fundamentally change how automotive systems are developed; from reducing development cycles by thousands of hours, to intelligently managing huge volumes of sensor data.
This closer integration between AI and in-vehicle sensing is set to increase consumer engagement, optimise business models and unlock a wealth of new use cases.
In this article we look at some of the latest AI research and discuss the potential it has to revolutionise the automotive industry.
Accelerating automotive development
As part of our ongoing AI research, we have invested in emerging technologies that have the potential to significantly reduce the burden on car manufacturers or their suppliers to deliver reliable systems for self-driving vehicles.
Generative Adversarial Networks (GANs) are an example of this; they utilise a configuration whereby two or more AI agents are locked in a competitive battle to continuously outsmart each other.
Vincent - our ground-breaking AI system - used a GAN that we finely tuned to sample just 8,000 works from the Renaissance period to the current day so it could build an understanding of where contrast, colour and texture change. Now trained, Vincent can take a simple human sketch and use its understanding to produce a beautiful completed artwork, reminiscent of the world’s most celebrated artists.
In an automotive context this means vehicle car manufacturers could utilise GANs to reduce their reliance on many thousands of hours of drive testing and instead augment limited drive journey training datasets.
For self-driving cars, AI experts could pull upon alternative sources of journey data such as simulators, games and dash camera footage, all of which will result in expert self-driving AI technology.
AI innovation in automotive will no longer be limited by the size or quality of the dataset. This will free AI experts to focus on optimising reliability and performance of self-driving solutions.
Using AI to challenge existing technologies
What if we could use AI to test limits of capability of an automotive electronic control unit (ECG) or sub-system? We believe it can.
We tested a well-regarded standard image classifier with input data that had been modified by an AI in ways which are almost imperceptible to a human.
These small changes, as you can see from the images below, have completely thrown off the standard image classifier that has struggled to identify the vehicle correctly. Using AI like this to test reliability of an ECU or subsystem will become routine and lead to better performing systems that increase our confidence and trust in automotive technology.
Original image classification (top 5 results):
- Police van
- Moving van
Adversarially-modified image classification (top 5):
- CD player
Making sense of the data avalanche
A recipe for responsible AI
Vehicles with self-driving technology must be capable of handling huge quantities of sensor data from numerous subsystems, including LIDAR, Radar, Ultrasonics and Cameras. These will create unprecedented volumes of data that cannot be managed by conventional means - AI is the only solution to manage this completely.
This data will not only be used to aid self-driving decisions in real time, but offer valuable experience information for training the next generation of self-driving vehicles. We see numerous opportunities where AI can better manage these high volumes of data. By using AI, data management strategies will be able to identify important driving events for transmission to vehicle car manufacturer’s servers and use intelligent compression techniques to optimise in car storage where finite resources are available.
The vehicle cockpit, which is rich in interfaces, will offer a growing opportunity for car manufacturers to differentiate by leveraging new sensing and AI technologies.
Not only does this offer the opportunity to use voice driven personal assistance, like Alexa or Google Home, but provide valuable health and safety insight. For example, cockpit sensors and cameras will have the ability to monitor occupant’s wellbeing to deliver actionable insight directly to the occupants, or perhaps healthcare providers.
In summary, AI offers huge potential to revolutionise the automotive industry. This is not just limited to enabling autonomous vehicles, but creating optimised development processes for the entire supply chain. The tight integration between sensor and AI systems also has the potential to offer significant value for the public and enable car manufacturers to gain a competitive advantage in a quickly evolving market place.
AI in Detroit
We'll be discussing all these topics and more, alongside our sister company Synapse, at the forthcoming TU Automotive event (6 – 7 June). We hope to see you there.