AI at the Wheel: How Artificial Intelligence is Driving the Evolution of Autonomous Vehicles

While about 200 years back, high-quality vehicles with unique features were unimaginable to a regular person, we have come a long way since then, with electric and hybrid vehicles becoming a daily part of our lives. 

Today, autonomous vehicles (AVs) are leading the innovation in automobiles that has entered the mainstream with much spectacle and expectations. But what is it, and how are they changing the face of vehicles? Let’s see!

A Look at Automation in Automobiles 

Autonomous vehicles (AV) are the kind of vehicles that use technology to partially or completely replace the human driver and drive itself to a predetermined destination in “autopilot” mode. At the same time, these AVs respond to traffic conditions, avoid road hazards, and provide more safety.

Different kinds of technologies used by these vehicles include sensors, lasers, radar, adaptive cruise control, active steering, anti-lock braking systems, and GPS navigation tech.

According to the Society of Automotive Engineers (SAE), there are six levels of autonomous vehicles based on human intervention. This classification, which is also used by the US National Highway Traffic Safety Administration (NHTSA), is as follows:

Level 0: The vehicle has no control over its operation, with the human driver doing all of the driving.
Level 1: The vehicle’s advanced driver assistance system (ADAS) comes with the ability to support the driver with steering and braking.
Level 2: The vehicle’s ADAS oversees accelerating and braking in some conditions, although the human driver is required to perform necessary tasks and pay complete attention to the environment throughout their journey.
Level 3: The vehicle’s ADAS can perform all parts of the driving task in some conditions, but when necessary, the human driver has to take control. This level of autonomy is currently achieved by AVs.
Level 4: The advanced driver assistance system of the vehicle can carry out all the tasks without needing human attention or assistance in certain conditions.
Level 5: The ADAS of the vehicle can do absolutely all tasks related to driving and in all conditions with no driving assistance needed from the human driver. In this stage, full automation is achieved. 

Autonomous vehicles offer the benefit of convenience and improved quality of life. Moreover, the physically disabled and elderly are able to gain independence. There’s also the potential to reduce traffic congestion, cut transportation costs, free up parking lots, and lower CO2 emissions dramatically. 

However, despite all the hullabaloo around autonomous vehicles, they haven’t yet achieved the success they were expected to see. So what’s the problem here?

Challenges Facing Autonomous Vehicles

Autonomous or driverless vehicles have fueled billions of dollars in investment in recent years, but there have been many pushbacks in AV launches as well as delayed customer adoption. So, let’s take a look at some of the most prominent challenges these vehicles are facing. 

Complex Driving Environment 

The systems used by AVs to track road signs, traffic signals, and the movement of objects on the road are not foolproof. They especially fail to understand real-world scenarios. 

For instance, if a flock of birds is sitting on the road, human drivers understand that birds will fly away as the vehicle moves forward, but AVs will either stop unnecessarily or slam the brakes. AVs also fail to detect complex social interactions like the hand movement or eye contact of another driver, signaling you to go ahead. 

Not to mention, AVs, as of yet, cannot behave safely when there are no traffic signs on the road. This means AVs can’t yet operate with maximum accuracy in any location in different countries. 

If a passenger wishes to visit a location not included in the map system, then he may also find it really difficult as the AVs may become disoriented. This calls for the need to have complex three-dimensional (3D) route maps to guide the vehicle, which is a time-consuming process if one wants to achieve coverage and accuracy.

Bad Weather

A big challenge for AVs is bad weather. These vehicles use a wide range of sensors: cameras to help view and identify the object, laser to track the distance of it, and radar to measure the object’s speed and the direction of its movement. 

Once data has been captured, the system makes a decision, but snow, fog, or heavy rain makes it difficult for the sensors to function correctly. So, adverse weather negatively impacts the accuracy of AVs’ sensing capability, which may compromise user safety. Then, there are issues of heavy precipitation and substances like water, oil, ice, or debris obscuring the lane markings.

Cost

Another big problem with AVs is cost; sensors used by these vehicles, like Lidar and radar, are expensive. On top of it, Lidar is still trying to strike the right balance between range and resolution. So, this presents the question: if several AVs are driving on the same road, does it cause their lidar signals to hinder each other?  

Liability

Yet another big question with AVs is that of accident liability; who is responsible for accidents caused by AVs? This will hold even more significance in regards to the fully autonomous level AVs that won’t have a steering wheel for a human to take control of the vehicle in an emergency. Then, there is insurance, which is another murky area for these vehicles.

Laws & Regulations

Despite AVs getting into the mainstream, laws and regulations surrounding them are still few and far between. Recently, the regulatory process for AVs in the US has shifted from federal guidance to state-by-state mandates. 

To prevent the rise of “zombie cars,” some states have even proposed a per-mile tax on them. Lawmakers have also written bills proposing that all AVs must have a panic button installed. 

Cybersecurity 

Given the highly connected transport system and the rollout of 5G, data privacy and cybersecurity are other problems with these vehicles. For instance, in 2015, Fiat Chrysler recalled 1.4 million of its vehicles to fix bugs as they could be hacked and remotely controlled. AVs need to ensure that they not only do not infringe on the consumer’s data privacy but must also protect the data from hackers. 

Infrastructure 

In order to bring AVs to the road, massive investments need to be made in infrastructure. AVs often need clear lane striping, places to store the data, and a more robust charging network. This will impact the city’s budget. Hence, there needs to be dialogue for public investments as well as community and industry outreach to expand existing infrastructure.

AI Paving the Way Forward for Autonomous Vehicles

In the face of all these challenges, artificial intelligence (AI) is taking charge and paving the way forward for autonomous vehicles.

The thing is the auto industry has been advancing rapidly in recent years with the advent of new technologies. AI is one such technology helping the automotive industry transform. Essentially, AI is all about making machines more intelligent. It involves the simulation of human intelligence in machines in order to make them think and act like us humans.

AI enables vehicles to recognize objects, predict what might happen next, even react to unexpected situations, and be better than human drivers at driving in complex traffic situations. According to Statista, the global automotive AI market is expected to reach a market size of $74.5 bln. 

As per the NHTSA’s study, human errors like impaired vision and he­aring cause about 93% of road accidents. Utilizing AI in AVs in the form of sensors and algorithms can allow for safer and more secure means of transportation, which can significantly reduce the casualties caused by human errors. AI’s ability to learn the environment and then adapt makes the technology more proficient in handling complex roads and situations.

AI is being used in AVs in a number of ways:

• The tech can help AVs predict the conduct of other drivers and pedestrians by equipping the vehicle with the ability to utilize analytics, predict any problems, and then prevent them from happening.
• By making use of machine learning, under which a model is trained on labeled datasets to map inputs to outputs correctly, AIs can help AVs with object recognition and modeling. Meanwhile, a model trained on unlabeled datasets can help AVs with anomaly detection, understanding intricate situations, and feature extraction.
• AVs rely on sensors like cameras, Lidar, radar, and ultrasonic sensors to obtain info about their surroundings. Here, AI algorithms can analyze this data to generate detailed maps, allowing AVs to make informed decisions.
• By leveraging Natural Language Processing (NLP), AVs can use voice recognition to interact with passengers. This way, AI can help vehicles understand human inquiries and respond effectively. 
• By enabling on-spot decisions based on real-time sensor data, AI helps AVs decide if the best response is to slow down or stop. This way, AI helps AVs with dangerous situations where humans are prone to making errors. The tech does it by analyzing data feeds across its sensors. It actually performs much better at cross-traffic detection, active monitoring of blind spots, syncing with traffic signals, and emergency controlling the vehicle.

Overall, AI in AVs can help collect data in real time, detect and identify objects, optimize trajectory, navigate road conditions, and predict failures. All these use cases of AI help autonomous vehicles achieve reduced traffic, accelerated energy-saving, improved accessibility, enhanced efficiency, and increased safety. 

Already, the technology is being used by automakers all over the world. For instance, Tesla’s Autopilot drove more than 3 billion miles in this mode in almost a decade. Elon Musk’s Tesla leverages sophisticated AI algorithms for accurate control. 

Waymo is another one that uses an AI-based self-driving system for complex route planning and intelligent reactions to its environment. The company has tested its vehicles by driving tens of billions of miles in simulation.

Daimler’s digital assistant, Audi’s R10 e-tron SUV, and Mercedes-Benz’s EQR4 autonomous driving system are some other examples. Other major contributors to AI in AVs include BMW, GM, Nissan, Uber, Volvo, Bosch, Mobileye, Valeo, Continental, Velodyne, Nvidia, and Ford.

Click here to learn how 2023 was the breakthrough year for AI and what can we expect going forward. 

Most Important AI Breakthroughs in the Autonomous Vehicles Space

2023 was a great year for advancements in AI, which made an impact on everything from art, finance, healthcare, and education to climate change, research, funding, and AVs. So, let’s take a look at some of the most important AI breakthroughs of 2023 in the autonomous vehicles space.

Most recently, researchers from Korea’s Incheon National University (INU) developed a novel end-to-end 3D object detection system, which is deep learning-based and Internet-of-Things-enabled. This system gives AVs improved detection capabilities, even under unfavorable conditions. 

Addressing the difficulty of sensors like cameras, lidars, and radars being vulnerable to obstructions, weather, and disorganized roads, this study adapted the YOLOv3 (You Only Look Once) algorithm to identify 3D objects by incorporating IoT as it enables objects to exchange data and communicate over the internet. 

The proposed system is designed to process RGB pictures and point cloud data as its inputs. It then outputs bounding boxes that are rated and labeled for identifying obstacles. This system is adept at detecting a diverse array of items and is capable of managing variations in both scale and rotation.

The study tested the system using the Lyft dataset and found that it demonstrated higher accuracy and lower latency. According to the team, the versatility of the proposed system extends beyond autonomous vehicles, finding potential applications in surveillance, robotics, and gaming as well.

Another project, Helm.ai, made an AI breakthrough that predicts driver intent and plans optimal paths. The company that creates AI software for the automation of robots and vehicles announced that this will allow Helm.ai to have scalable L2/L3 and L4 deployments. 

The company’s DNN-based foundation models are trained using its proprietary tech, Deep Teaching, which utilizes real driving data for complex driving environments. 

Now, its model also analyzes surrounding vehicles and pedestrians to accurately predict their likely actions in diverse urban situations and, based on that, generate the most efficient and safest path for the AVs to take. The company’s platform works with different hardware configurations in a seamless manner and enables efficient training and validation.

“Our software platform addresses the critical perception challenges of urban environments, paving the way for scalable development and validation of AI-powered intent prediction and path planning.” 

– Vladislav Voroninski, Helm.ai’s CEO

This year, the electric vehicle pioneer Tesla also made advancements in its Full Self-Driving (FSD) software. Its latest version, 12 (v12), allows the company to get another step closer to achieving Level 4 or Level 5 autonomy with its cars. 

In August, Musk demonstrated FSD v12 driving the vehicle autonomously and performing tasks like parallel parking, obeying traffic lights, and navigating roundabouts. What sets this version apart from its previous ones is the heavy dependence of FSD v12 on AI-powered self-training neural networks. 

What this means is that instead of requiring human programmers to hard-code responses for different driving scenarios, AI will analyze tons of data collected from Tesla’s vehicles and will then choose the most appropriate response.

This development takes Tesla one step closer to achieving its broad goal of a robotaxi business, which, according to Ark Invest, even in a bearish scenario, generates $200 billion (over $600 bln as per its most optimistic projection) in annual revenue.

Earlier this year, another breakthrough came for AVs in the form of a camera imaging system, HADAR, or ‘heat-assisted detection and ranging.‘ Researchers from Michigan State University and Purdue University utilized AI to develop HADAR, which interprets heat signatures to provide detailed and sharp images while cutting through the optical clutter. 

Their AI model leveraged machine learning algorithms that gather data from commercial infrared cameras to recognize the physical properties of objects and their surroundings, allowing HADAR to reconstruct clear night-time scenes. 

Given that the system can detect thermal radiation patterns, material formation, and temperature with great success, it has vast potential, including contactless public security screenings and even overcoming the fear of the dark. However, HADAR has challenges in terms of equipment cost and the need for real-time calibration.

Ford Motor Company also created a wholly owned subsidiary called Latitude AI to develop a hands-free, eyes-off-the-road automated driving system. The automobile giant has already got over 50 million miles of hands-free driving in its Ford BlueCruise. 

Now, with Latitude, the idea is to automate driving tedious, stressful, and unpleasant times such as long stretches of highway or bumper-to-bumper traffic. Speaking on automated driving, Ford’s chief technology officer, Doug Field, said:

“We see automated driving technology as an opportunity to redefine the relationship between people and their vehicles.” 

Concluding Thoughts

So, as we saw, depending on the level of human help needed, autonomous vehicles fall under different categories, viz. automation for driver assistance, partially automated driving, highly automated driving, fully automated driving, and completely automated vehicle. With the advent of AI, the possibility of AVs finally reaching their final stages is closer than ever.

The future of the AI market in the automotive industry is clearly promising. It was standing at over $6 billion in 2022 and is projected to grow at a CAGR of 55% by 2032. 

Advances in AI algorithms, such as sensor technologies, computing power, and predictive maintenance solutions, will further help autonomous vehicles address their challenges and gain mainstream adoption!

Click here to learn all about investing in artificial intelligence.