The rise of artificial intelligence has been brewing for decades. In the past year alone, we’ve seen incredible developments in several forms of AI. From language models to personal assistants, this new technology has the potential to transform countless industries – and in some companies, it already has.
In the world of autonomous vehicles and self-driving semi-truck companies, Torc uses artificial intelligence every day as a tool for those self-driving semi-trucks, and to make some of our internal processes more efficient. AI is a part of our autonomous driving systems, but there’s a ton of misleading information out there about how it all works. Does artificial intelligence make all the same decisions that a human driver would make? What companies make the AI for self-driving cars and trucks? With the help of Torc’s Director of Engineering in Product Development, Justin Brown, we’re answering all these questions and more in this self-driving AI deep dive.
How do self-driving vehicles use AI?
Autonomous trucks collect an incredible amount of data while they are on the road due to the multitude of sensors used to localize and understand the environment around the vehicle.
Some of this data is fed into a machine learning pipeline, which is both a subset of artificial intelligence and a broad term used to refer to automated data processing.
Torc combines AI and traditional approaches to data processing to ensure it is balanced and traceable. On the road data processing enables the software in a self-driving 18-wheeler to take in data from its sensor suite and automatically classify it. Artificial Intelligence shines particularly in vision-based classification, helping the system define signs or objects made for human vision. It’s what allows our trucks to “see” a traffic light turning yellow, process that change, and decide to slow down.
At their core, self-driving semi-trucks are decision-making machines. Artificial intelligence helps provide additional information to make those decisions.
Think about the way that you learned to drive. There’s a good chance you made some mistakes that taught you a ton about putting the pedal to the metal. There’s also a good chance that you learned a lot about why being a predictable driver is such an important contender for safety – and you might’ve learned that human drivers aren’t always predictable.
Artificial intelligence and machine learning allow self-driving software systems to learn about the environment just like people do. Subsets of artificial intelligence, like the deep learning algorithms used in autonomous driving technology, run data through several layers of neural network systems (computer systems modeled after the human brain and nervous system – complete with artificial neurons) in order for the overall system to “learn by example”. The system may be trained on thousands of types of traffic lights, for instance, so the autonomous driving system can recognize one on the road.
Machine Learning and Responsible Use
While machine learning is certainly an important tool in our toolbox, it’s not the only approach we use for development. Our autonomous driving system uses diverse technologies, each selected for the right use case. Together, traditional code and artificial intelligence come together to recognize patterns, predict the movement of traffic and influence driving behaviors.
We also uphold rigorous standards of ethics and safety when it comes to machine learning. It is important to ensure our data collection processes are explainable and transparent. We work alongside government officials and regulators to create guidelines around concepts like data bias, sustainability, and other key factors. At Torc, we carry out careful consideration of issues like fairness, accountability, transparency, and privacy, in each stage of our development. As the technology continues to advance, we’ll work alongside our colleagues, policymakers, and researchers to create mindful processes around autonomous driving systems.
What kind of AI do self-driving vehicles use?
With the rise of chatbots, language models, and even neural networks that can create images, AI has become a vast and sprawling form of technology. However, there are many different types that can be used depending on use case. Currently, computer science defines AI in four types: reactive, limited memory, theory of mind, and “self-aware”.
Reactive AI is the most basic form of AI, wherein a machine is programmed with a specific output based on the given input. Most of us encounter reactive AI in things like recommended shopping algorithms, streaming service recommendations, and spam filters.
Theory of mind AI, or AI, is just coming to fruition; this type of AI interacts with the emotions and thoughts of people. Machines equipped with theory of mind AI will be able to gauge a person’s facial expression and adjust behavior based on that calculation.
Self-aware AI is the science fiction take on artificial intelligence. These robots of the future are entirely free-thinking, complete with human-level consciousness and intelligence. Unfortunately for our movie and TV show sets, these machines don’t yet exist.
Lastly, limited memory AI is the most common form of AI used today and, therefore, the most established – especially when it comes to self-driving vehicles and autonomous driving technology. It uses both historical and observational data alongside pre-programmed information to make predictions about the world around it.
What’s the difference between other forms of AI and ADS AI?
Limited memory AI is used in autonomous trucks, sometimes referred to as AV trucks, and other forms of self-driving vehicle technology for precisely the same reason as those chatbots. Limited memory AI allows a system to make decisions on its own, but within designated parameters.
In an autonomous 18-wheeler, this would be most apparent in things like road obstacles. Say that an automated semi-truck is driving along the interstate in Arizona. As the truck drives on, it detects a traffic cone knocked over in the lane ahead. The vehicle perceives the item and, via those pre-programmed parameters, calculates that this is an object that has the ability to move, but likely will not. Those pre-programmed parameters also inform the truck that it’s not necessarily a dangerous object, but it’s best to avoid all potential obstacles where possible.
The autonomous driving system for the semi-truck is using this prior information to “remember” that the cone can move. It can also be a signal to highway users that there’s reason for caution surrounding the cone. Via these deductions, it uses behavioral parameters to determine the best course of action. In this case, it’s likely moving out of the lane until the truck is clear of the obstacle.
Like any other form of AI, the artificial intelligence used by autonomous 18-wheelers is rife with myths and misinformation. “There’s a common misconception that the vehicle’s processing, and its AI systems, happens off the vehicle,” says Justin. “But navigation, perceiving what objects are and how fast they’re going – all that happens on the vehicle, so it doesn’t require an internet connection or anything like that.”
As a self-driving truck company, Torc takes pride in the safety of our technology and processes. Like all of our sensors and software components, we utilize artificial intelligence with the safety and security of pedestrians, drivers, and other highway users in mind.