AI in transportation: How companies are building an autonomous future

March 15, 2022

2022 Blog Insights Research Research 2022

By Jonathan Woods
March 15, 2022

Lift the hood on the transportation industry, and you’ll find a rush of AI research and deployment underway. Spurred by potential efficiencies and customer desires for better performance, companies are using AI systems to optimize manufacturing, enhance driver assistance models, and collect the data required for a future featuring increasingly autonomous transportation systems.

The AI innovation occurring in the mobility sector is remarkable, and advances made by Vector Industry Sponsor companies – Thales and Linamar – illustrate this. Both are devoting serious attention to AI, and recently participated in Vector’s Computer Vision project, a collaborative industry project that brings Vector sponsor companies together with AI researchers to apply state-of-the-art models to real industry use cases. The work of these two companies, and their projects relating to anomaly detection and autonomous mobility systems, provide a window into how AI is transforming the transportation industry and how companies are setting themselves up for an AI-driven future.

Quality assurance:  Identifying defects at scale

For Linamar, AI can increase manufacturing efficiency by automating quality assurance. Integration Engineer Tristan Trim described how at Vector’s Computer Vision Symposium:

“At Linamar, we do a lot of precision manufacturing. … But as you can imagine, every little bit of attention to detail added to each part across 10,000 parts a day is going to start adding up. … What would be great is if we could find some kind of automated way of detecting defects in our manufactured products without having to give so much focus there.”

In the Computer Vision project, Trim was able to do just that. Trim trained models to detect anomalies in images of objects like transmission components, and to then reproduce those images with any imperfections highlighted. Using a dataset of 5000 high-resolution images, the model learned to identify any abnormal pixels. These pixel abnormalities indicated defects and pinpointed exactly where they were. This work shows that it’s possible to develop an automated system that can detect any unique defect, enabling accurate quality assurance at scale.

Toward autonomous trains: Detecting vehicles, obstacles, and trackside workers

At Thales, requests for systems “that perform as well as or better than a human” have led the company to explore how computer vision can support the long-term development of autonomous trains, says Veronica Marin, Manager of Advanced Algorithms, Research & Technology, Transportation Systems.

“We’re bringing in autonomy functionalities to the train,” Marin says. “That means the trains will be kind of agents.” In other words, systems on trains will need to figure out where they are, what’s in their environment, and what actions to take based on that information.

Unlike driverless trains, which have been around some time, autonomous trains won’t rely on centralized and wayside infrastructure to determine their locations relative to other vehicles or to guide their actions. Autonomous trains will need to act and interact independently, meaning systems on the trains themselves must be constantly aware of their environments and positions within them. That requires computer vision models that can accurately and in real time detect and classify objects, like vehicles, trackside workers, or obstacles.

For now, Thales’ computer vision models are geared to provide driver assistance only ― a necessary step on the way to full autonomy.

Making that transition involves another activity that transportation companies are focusing on today:  gathering the data to train their systems. Currently, such rich datasets don’t exist – but they’re crucial for autonomous ambitions.

Forward-looking companies like Thales are not simply waiting for such datasets to emerge. They’re being proactive, and using their current systems to collect the data they’ll need to train for future models. Marin says, “With that data, we are building the confidence for a safety certification.”

Computer vision is only one of several AI techniques transforming the transportation sector. AI is on the rise throughout the industry, and proactive companies are exploring the new capabilities and operational efficiencies various AI techniques can offer.

One way to remain at the leading edge is to become a Vector Industry Sponsor. The Vector Institute provides support for companies in the transportation sector in the form of expertise, applied industry projects featuring cutting-edge AI research, and events focused on AI and mobility. Vector Faculty member, Angela Schoellig will lead an upcoming series on Autonomous Navigation Systems and Data-led Control Systems – exploring reinforcement learning for robotics.

Speak with Vector’s Industry Innovation team to learn how Vector sponsorship and collaboration can accelerate AI innovation and technology transfer today to develop leading, ‘made-in-Canada’ products and solutions for the future.

Related:

2024
Machine Learning
Research
Research 2024

Vector Institute researchers reconvene for the second edition of the Machine Learning Privacy and Security Workshop

Headshot of Vector Faculty Member Wenhu Chen
2024
Insights
Research
Research 2024

Vector researcher Wenhu Chen on improving and benchmarking foundation models

2024
Research
Research 2024

Vector Researchers present papers at ACL 2024