Imagination Technologies and Vivacity Labs on AI and cars

Imagination Technologies and Vivacity Labs discuss the considerations that need to be made concerning AI and ML for autonomous vehicles to be successful

Autonomous vehicles, or self-driving cars, are being developed and tested by a number of manufacturers across the globe. These vehicles use a combination of sensors, cameras and artificial intelligence (AI) to travel between destinations.

Research has shown that the production of self-driving cars is expected to reach 800,000 units worldwide between the years 2023-2030.

Despite concerns around the technology and its ability to safeguard passengers from harm, KPMG has predicted the adoption of self-driving vehicle technology could reduce the frequency of accidents by approximately 90%.

Now, as the technology matures and autonomous cars become commercially available, technologists and car manufacturers need to dispel concerns around the ethics and safety of self-driving technology.

Noting the benefits of this technology, Gilberto Rodriguez, Director of Product Management - Artificial Intelligence and Ethernet Connectivity at Imagination Technologies said: “One benefit would be the ability to reduce the need for drivers. This can significantly reduce the cost per mile of driving and, with it, the need to own a car. Robotaxis will eventually become extremely cheap to use, making car ownership questionable for most people in towns and cities. With more people travelling and more goods being delivered to the doorstep, the technology will alleviate backlogs.”

“Secondly, with smart mobility we can get better use out of vehicles, with models like Car-as-a-Service rolling out in the future. This unlocks better prediction, control and management of traffic. This can bring significant improvements such as adding flexibility to deliver goods outside rush-hour and managing car fleets according to actual demand – further improving the overall traffic experience,” he continued.

Adding to this, driverless technology has the ability to increase traffic efficiency by reducing network congestion as the technology can communicate with each other and react to real-time events and flow options.

Mark Nicholson, CEO and Co-Founder of Vivacity Labs comments that the core benefits of this technology are safety and convenience: “With human error, the cause of 94% of crashes, self-driving cars reduce the risk of misjudgement and making mistakes that can result in negative outcomes.”

Neural networks feeding AI and ML algorithms for autonomous cars

When developing this technology, car manufacturers use large amounts of data from image recognition systems along with machine learning (ML) and neural networks.

Neural networks are key for this technology as they identify patterns in the data to feed the machine learning algorithms.

“The richness of these datasets is such that traditional technologies (created before neural networks) can't cope. Simple heuristics (rules saying "if this, then that") can't cope with the messiness of real life. This is why neural networks are very good at encoding this messiness but need the training to do so,” said Nicholson.

“So the top autonomous vehicles now have billions of hours of simulated training under their belts, and millions of hours of real-world training, in order to give the neural networks the material they need to encode real-world messiness,” he added.

As with any AI technology, datasets are imperative to train deep learning models and achieve high levels of accuracy. 

With self-driving vehicles, Rodriguez explains: “The data needed comes from information captured from surroundings. This can be dynamic coming from sensors such as Radar, LIDAR, cameras, Infrared or static from 3D maps in combination with GPS. As all sensor data is processed, the car is placed into a virtual replica of the world, giving it a frame of reference for positioning and for what is happening dynamically around it (cars, trucks, motorcycles, pedestrians). AI will help work out the safest path through the road using path planning algorithms. This is then converted into controls for the car’s acceleration, brakes, steering and so on.”

Overcoming challenges with AI and ML to fully commercialise autonomous technology

Although many companies are developing and testing driverless technology, including Audi, Tesla, Google and BMW, there are still significant technical challenges the industry will face.

Testing these systems is crucial to the development of the technology and algorithms within autonomous vehicles. Bringing a financial challenge to the table, some believe technology and automotive companies could spend up to $10 billion to test and perfect their technology.

With the huge expense that comes with developing this technology, Rodriguez says it is important that “OEMs understand where they fit in the value chain. Driverless car technology can go from capturing data to full validation/certification of a vehicle.”

“With the neural network accelerator performing most of the computer, it’s important to pick the right one. However, machine learning is evolving very quickly, so flexible programmable hardware is also a requirement,” he added.

Despite these challenges, Nicholson believes this technology is ready to be rolled out, although it may still be a number of years until it has the ability to replace humans in all different types of routes. 

He said: “The technology is mature and ready for commercial use but just not in the way we would imagine for normal cars. The technology is 99.9% there - but the question is always how many 9s do we need to add to be confident in a safety-critical situation?” 

“They've been deployed in small segregated areas; they're now operating in large scale parts of cities, but the cars still require exhaustive testing in that city before they can be unleashed, so they can't be deployed just anywhere yet. We're still a number of years away from a world where you can enter a destination hundreds of miles away and your car will take you there.”

Ethics and misconceptions with AI and self-driving technology: what needs to be considered?

Within all applications of AI, be it AI in healthcare, military operations or facial recognition, bias and ethics is and will remain a huge issue.

Outlining these ethical biases, Nicholson stated there is a number that need to be considered:

  • “Avoid visual bias, such as whether it can spot a person of colour just as well as a white person or a woman compared to a man
  • Avoiding technological bias, such as whether it can see a cyclist because they have bought a device that broadcasts the cyclist's location, but the cyclist who can't afford that device has a higher chance of being hit
  • Decision-making bias is also crucial, with questions around whether there’s a 10% chance of hitting a male or a 10% chance of hitting a female, making sure that there isn't any bias in the choice of action
  • Finally, self-preferential bias such as if a car is a Google car and it gives way to a fellow Google car as opposed to a Ford vehicle at the next junction.”

Misconceptions in the media mean that many people see AI as a threat to humankind. To overcome these misconceptions and fears within autonomous vehicles, Rodriguez believes the narrative should be more open and testing needs to remain rigorous: “It is important that the technology is proven under controlled environments before we deploy driverless solutions globally. Each road across every city and country has its own specific nuances which bring a variety of navigation challenges.” 

He concluded: “One of the key elements will be acceptance of the technology.  This may be somewhat generational as the younger generation, being less car-centric, are happy to adopt new tech that makes their lives easier and less costly in towns and cities. Why own a car when it is so easy and cheap to call up a driverless vehicle which is guaranteed to be safer than a human-operated vehicle?”

Share

Featured Articles

Pick N Pay’s Leon Van Niekerk: Evaluating Enterprise AI

We spoke with Pick N Pay Head of Testing Leon Van Niekerk at OpenText World Europe 2024 about its partnership with OpenText and how it plans to use AI

AI Agenda at Paris 2024: Revolutionising the Olympic Games

We attended the IOC Olympic AI Agenda Launch for Olympic Games Paris 2024 to learn about its AI strategy and enterprise partnerships to transform sports

Who is Gurdeep Singh Pall? Qualtrics’ AI Strategy President

Qualtrics has appointed Microsoft veteran Gurdeep Singh Pall as its new President of AI Strategy to transform the company’s AI offerings for customers

Should Tech Leaders be Concerned About the Power of AI?

Technology

Andrew Ng Joins Amazon Board to Support Enterprise AI

Machine Learning

GPT-4 Turbo: OpenAI Enhances ChatGPT AI Model for Developers

Machine Learning