February 18, 2021
AI in automotive: its present and future
Today’s consumers expect their cars to become as smart and personalized as their mobile devices and personal voice assistants. In combination with AI, many emerging technologies and concepts including IoT data analytics, cloud computing, 5G, and big data open up completely new horizons for interconnectivity between vehicles and the surrounding environment.
Notably, the automotive industry has always been proactive in the adoption of emerging technologies. Now, companies in the automotive sector are increasingly integrating AI to tap into connected vehicle technologies, autonomous driving, and predictive maintenance.
Currently, the AI development rate is exponential. According to Deloitte’s research, the global automotive AI market is projected to reach almost 27 billion US dollars in 2025.
One of the most widespread and demanded use cases for AI in automotive is advanced driver-assistance systems (ADAS). According to the World Health Organization, 1.35 million people lose their lives yearly because of road traffic accidents. While inadequately designed road infrastructures and traffic laws need to be continuously addressed, human error is still among the most recurring reasons for traffic crashes. At this point, turning to machine learning development to reduce the probability of human errors resulting in adverse events is the logical next step.
The National Highway Traffic Safety Administration (NHTSA) defines six levels of driving autonomy, as shown in the image below. Currently, most AI-powered vehicles range from level 1 to 3 on the NHTSA scale. In order to make a car fully autonomous, meaning that a driver can essentially become a passenger and entrust all of the driving to the machine, an immense amount of training needs to be done.
Adequately understanding all driving nuances and contingencies requires lavish amounts of data to be captured and processed by a multitude of autonomous vehicle sensors in real time. Computer vision-enabled cameras make sense of traffic signals and street signs, as well as recognizing nearby vehicles and pedestrians. Lidar sensors gauge the distance between vehicles and other surrounding objects to automatically command the car to slow down or speed up.
Interestingly enough, Tesla, which is a definitive leader of the autonomous car manufacturing industry, is all against the use of Lidar as this technology can’t provide AI systems with sufficient information about an object. Whether it’s a plastic bag or a dog, Lidar will define them both as just moving objects.
When it comes to fully autonomous vehicles, the aforementioned data sources is just the tip of the iceberg. Experts claim that a vehicle needs to process more than 1 TB of data per second on average to become fully autonomous. Currently, there are no other systems that require such an enormous computing demand to function properly. Companies are turning to various cloud solutions, where the data is labeled, processed, and used to optimize these algorithms. However, it can be argued that the mass deployment of autonomous vehicles doesn’t align with the current trend of low energy consumption and poses certain economical hurdles.
Although there are already somewhat successful examples of driverless vehicles, most of them were extensively trained on the same route. This allows an AI system to develop a highly detailed map that accounts for every nuance of the road. Admittedly, this is not a practical solution for deploying autonomous vehicles on a larger scale, as considering every possible route even in one region is not feasible.
At the end of 2020, Tesla announced that the beta version of its full self-driving (FSD) software will be available to a very narrow range of selected drivers. At the end of January 2021, the FSD beta has managed to impress the public by handling some of the most complex routes in Berkley, CA, hinting that Level 5 autonomy is not as far as it may seem.
However, it’s critical to note that regardless of the FSD’s finesse, every driver on the road today needs to be prepared to take over the control of the vehicle. Viral videos of people sleeping or looking at their phone while on the road creates a very misguided impression about AI-assisted driving.
With the rapid advancements of IoT use cases, AI, access to 5G networks, and cloud computing, vehicles can be connected to each other, mobile devices, and infrastructures, enabling a higher level of safety and efficiency. For example, cars can ‘talk’ to each other to ensure they keep a safe distance. Connected delivery trucks can formulate tight convoys at the highest speeds, enabling minimal wind resistance, and, consequentially, decreasing fuel consumption.
In terms of safety, the technology behind connected vehicles is a key enabler of preventing crashes, rather than helping people to survive them. Connected cars can substantially reduce the number of fatal road accidents. While today’s cars often make use of smart sensors and computer-vision enabled cameras to alert drivers about potentially dangerous road situations, the connected vehicle technologies allow this to happen much earlier, providing drivers with considerably more time to react.
For example, if a car is speeding around the corner and is out of sight, the driver will still receive an alert. This isn’t possible with the conventional sensor-based tech. Most importantly, the connected vehicle technologies have all the potential to become widespread as the cost of implementation is much lower than that of radar or smart camera installments.
Furthermore, connected vehicles are posed to reduce traffic congestion and travel times. Traffic managers would be able to get a bigger picture of the road situation and more efficiently control traffic flow.
Considering the massive success of AI-powered voice-based personal assistants like Alexa and Siri in the last few years, car manufacturers are now integrating similar applications in their products. While some of the industry players decided to implement the aforementioned third-party personal assistants, others have chosen to create their own state-of-the-art voice-recognition software.
For example, Amazon has recently started collaborating with major automotive companies including Toyota to integrate Alexa into their infotainment systems. The assistant can now adjust the temperature, provide information about how much gas is left in the tank, make calls, change radio stations, etc. BMW, on the other hand, has decided to go all-in and integrate multiple assistants. Besides its own voice-recognition technology, the company is now leveraging Microsoft Cortana to take care of work-related applications, Amazon Alexa to entertain, and its own software to provide information about the car condition.
It’s noteworthy that most of these applications are personalized. The assistants can remember drivers’ preferred settings and suggest certain adjustments based on the context and user history. Some of the car manufacturers like Toyota go even further and integrate emotion recognition capabilities, allowing conversational AI to take into consideration the user’s emotional state based on their voice tone and facial data sourced from the dashcam.
AI is key to the success in predictive maintenance. With the myriad of connected sensors installed across vehicle components, maintenance guesswork can finally be eliminated and serious problems can be addressed before they escalate. For instance, smart sensors can alert drivers about a necessary oil change or low tire pressure. Besides car owners, predictive analytics also immensely benefits car manufacturers. By keeping a vehicle proactively monitored, drivers are more likely to turn to manufacturers for maintenance rather than to independent car repair shops.
In the logistics and transportation industries, downtime is one of the most fearsome issues. Predictive maintenance will become key to lessen the number of these situations by providing a real-time remote diagnosis of vehicles. Last year, Michelin has announced that all of its automotive tires will be equipped with RFID by 2023. While drivers will be able to see tire status on their dashboard themselves, Michelin explicitly points out that RFID-enabled tire monitoring is essential to making automated vehicles economically feasible.
With the rapid proliferation of autonomous vehicles, legislative matters become paramount. Most importantly, who needs to be held accountable in case of an accident? The answer to this question is a decisive factor in the widespread AV adoption. Currently, it all comes down to manufacturers working closely with governments and research organizations to create a supportive environment for technical development while ensuring safety.
Level 5 autonomy relies upon advanced AI algorithms including convolutional neural networks, which are still often complex to decipher. Not only it’s difficult to understand why exactly an embedded AI system makes certain decisions, but some of the ethical complexities also remain unsolved. Many fatal accidents can’t be avoided, implying that fully autonomous vehicles would need to make some of the toughest ethical decisions. For example, should an AI system decide to run over pedestrians to save the passengers or vice versa? For autonomous vehicles to become a reality, such issues need to be definitively resolved.
Existing regulative frameworks like GDPR also pose considerable limitations regarding the collection of personal data, which is inherited in AI-enabled driving. Most data processed by AI-assisted vehicles are considered personal data. For example, video footage collected from dashcams or geolocation data must be processed in line with the GDPR data protection rules.
Given that there is a myriad of functions that require personal data use, it’s often complex to adequately obtain users’ consent in the driving environment. Think about the irritation caused by pop-up windows that require your consent for personal data processing each time you enter a new website, and multiply it by 10. In addition, with the further advancements of AI-assisted driving, more sensors are getting deployed, hence more data is being processed. This can cause companies to collect way more data than they need, which is strictly prohibited by the GDPR. Moreover, some data collected by AI-assisted vehicles can be related to criminal convictions. For example, speeding is a criminal offense and such data can be processed only under the control of an official authority.
Fortunately, all the industry players including legislators and manufacturers are actively adjusting the relevant legal frameworks. In November 2020, the European Automobile Manufacturers Association released a paper that highlights the current legislative challenges and proposes a clear roadmap for overcoming them.
In a nutshell, with the increasing volumes of data needed for AI-assisted vehicles to operate, more security risks arise. Fraudsters’ motives range from vehicle tracking to data theft to vehicle ID reassignment. For example, in March 2019, the hacking duo Fluoroacetate hacked a Tesla Model 3 car via its embedded browser and thus won the Pwn2Own hacking contest.
One of the most effective preventive measures revolves around encryption-key management systems that are exclusive to each vehicle. In this case, encryption keys must be regularly updated. When data is transmitted to a third party, users’ consent has to be systematically obtained. While such mechanisms are not user friendly, drivers’ confidentiality comes first.
Currently, AI is a definitive key factor in the automotive industry proliferation. The inevitable shift from hardware to software in the automotive industry requires vehicle manufacturers to reimagine their workflows and pay close attention to relevant regulatory frameworks. These efforts combined make overcoming of the aforementioned legal, ethical and security challenges just a matter of time.
Learn how data storytelling helps businesses get all the teams on board for reaching common strategic goals.
Find out how companies harness voice marketing technologies to satisfy customers and boost employee productivity.
In the face of the growing uncertainty surrounding machine learning adoption in the banking industry, Itransition responds with an actionable framework.