Torc to Present Nine Papers at CVPR 2024

Torc to Present Nine Papers at CVPR 2024

Driving Autonomous AI with Torc’s Head of Intelligence

This summer, Torc’s Head of Artificial Intelligence, Felix Heide, and his team of both Princeton University and Torc colleagues will present an extraordinary total of nine papers at the IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR). Featuring information on geometry reconstruction, generative method, learning techniques, novel datasets and more, Felix and his team will exhibit their core competencies in researching and creating AI tools and techniques for self-driving artificial intelligence and beyond.

Torc at IEEE/CVF CVPR Conference

As one of the primary events in the field of computer vision, CVPR brings leading researchers, industry professionals, academics, and more, to a weeklong conversation about the latest findings in the world of computer vision and pattern recognition. With its main exhibition floor, presentations, and expert attendees, CVPR offers all levels of computer science professional with an opportunity to share their cutting-edge research and collaborate within the community. The IEEE/CVF CVPR Conference serves as the perfect backdrop for the innovative research Torc is spearheading.

The nine papers at CVPR 2024 cover a wide range of topics, each offering unique insights and contributions to the field. From advancements in neural rendering to new types of lidar, Felix, joined by his Princeton and Torc teams, has developed comprehensive content that reflects the depth and breadth of the industry’s expertise. Their work showcases the latest developments in computer vision and pattern recognition and indicates the ongoing progress of autonomous trucking technology.

Innovating Freight with AI

The significance of this research extends far beyond the confines of academic discourse. By pushing the boundaries of AI and machine learning, this work has the potential to revolutionize the autonomous driving industry and reshape the way we interact with technology. At Torc, these contributions help us pave the way for safer, more efficient freight systems that will benefit fleet professionals, everyday consumers, and all highway users.

As we move forward, AI research points us towards exciting new possibilities for collaboration and innovation. As academia, industry, and other research organizations come together to tackle the challenges of tomorrow, conferences like CVPR play a crucial role in fostering dialogue and driving progress. By sharing their insights and expertise, researchers like Felix and his team can shape the future of AI and propel us towards a world where driverless trucks enhance our lives in ways we’ve only begun to imagine.

These contributions not only highlight the cutting-edge research happening at Torc Robotics but also underscore the validity and potential of autonomous technology. As we continue to innovate our automated truck technology, our team’s contributions propel us to drive the future of freight in new and innovative ways. For our team, self-driving AI isn’t just a future concept, but a tangible reality that will make our world safer, faster, and stronger.

 

About Felix Heide, Torc’s Head of AI

Felix Heide is an industry leader in both artificial intelligence and autonomous technology. His journey to becoming an expert in general artificial intelligence, self-driving AI, and machine learning began long before his time at Torc, ranging from his first internship with NVIDIA Research in 2013 to his current role leading autonomous trucking development.

Felix is an Assistant Professor at Princeton University, Head of AI at Torc Robotics, and founder of the self-driving vehicle startup Algolux (now part of Torc). He is researching the theory and application of computational imaging and computer vision systems. Exploring imaging, vision, and display systems end-to-end, Felix’s work lies at the intersection of artificial intelligence, computer graphics, and computer vision.

He received his Ph.D. from the University of British Columbia, his undergraduate degree from the University of Siegen, and was a postdoc at Stanford University. His doctoral dissertation won the Alain Fournier Dissertation Award and the SIGGRAPH outstanding doctoral dissertation award. He won the NSF CAREER Award 2021 and the Sony Young Faculty Award 2021. He was named a Packard Fellow in 2022 and a Sloan Research Fellow in 2023. Felix was named SIGGRAPH New Significant Researcher in 2023.

 

 


This article is part one about Torc’s presence at CVPR this year. Look for more news about the CVPR event in June.

Lane-Keeping in Self-Driving Trucks: Precision and Trust

Lane-Keeping in Self-Driving Trucks: Precision and Trust

Across all the features that self-driving technology has to offer, we might think of lane keeping as one of the most basic features possible. However, lane keeping is a complex behavior that relies on multiple components, sensors, and procedures to complete safe driving behaviors. In self-driving technology, precision is everything, making lane keeping a foundational necessity that underscores the safe and efficient operation of robotic trucks.

What is lane keeping?

Lane keeping is a critical driving functionality that ensures a vehicle stays within its designated lane on the road. Many newer consumer cars and commercial semi-trucks have some form of autonomous lane-keeping system programmed in, albeit there are several nuances and differences between types of lane-keeping systems, such as:

Lane Keep Assist (LKA)

Lane Keep Assist is a feature that can be toggled on and off on most vehicles. It typically works via camera, allowing the LKA to “see” the lane lines and nudge your vehicle within the lane lines when it begins to drift. However, because it’s camera-based, this feature may struggle to perform in muddy, snowy, or especially rainy conditions.

Lane Keep Assist is sometimes confused with Lane Departure Warning, which alerts drivers via haptic feedback, audible alerts, and sometimes indicator lights, if they’ve begun drifting out of the lane. Unlike other forms of lane-keeping systems, lane departure warning won’t correct the vehicle’s path. Instead, its job is to inform the driver that the vehicle is exiting the lane.

Lane Centering Assist (LCA)

Lane centering, sometimes called autosteer, takes LKA a step further. This feature is typically part of a vehicle’s adaptive cruise control, wherein a vehicle performs most highway behaviors itself while under human supervision. Lane centering is an active technology that keeps a vehicle in the center of its lane and can typically be turned on and off.

Today, many Class 8 trucks come equipped with various forms of Lane Keep Assist, Lane Centering Assist, and Adaptive Cruise Control. Aside from making drives safer for truck drivers and other highway users alike, these features can adjust throttle inputs and gear ratios for more efficient driving behaviors. By utilizing these features, drivers can optimize the amount of fuel their equipment consumes, reducing costs across the board.

Why is lane keeping important?

Lane keeping plays a foundational role in ensuring the safe and efficient operation of any vehicle, including our autonomous trucks. By enabling our robotic trucks to steadfastly maintain their designated lanes, we’re not only addressing a core competency in highway navigation but proving that our technology can be a safe foundation for a self-driving system.

Lane-keeping is important because of its impact on safety, but there are a few other reasons why we focus on this behavior as one of the most integral in safe-driving technology.

lane keeping

 

Traffic Flow and Predictability

When vehicles stay within their designated lanes, it reduces the likelihood of swerving and abrupt lane changes. When we humans learn to drive, we learn that being a predictable driver means being a safe driver.

Our autonomous technology is no different. In order to reduce the likelihood of traffic accidents and promote smoother traffic flow, we must ensure that other drivers are able to anticipate what our self-driving truck is going to do at all times. Whether that means keeping a consistent, steady pace within one lane or using an indicator light to shift lanes, predictability should be at the forefront of all lane-keeping behaviors.

Efficiency

By staying within the given lane, vehicles maximize the use of available road space, allowing more efficient traffic flow. During peak hours, when space is at a premium, this can reduce delays, avoid bottlenecks, and make it easier for vehicles to travel at a consistent speed throughout their journey.

How do self-driving cars and trucks stay in their lanes?

Self-driving trucks and self-driving cars stay in their lanes via cameras, Global Navigation Satellite Systems, LiDAR (or Light Detection and Ranging sensors), and more. Thanks to the work of autonomous driving engineers, our self-driving semi-trucks’ perception suite can recognize lane markers, interpret them correctly, and communicate this information to the rest of the system. From there, the autonomous driving system can utilize the information to maintain a set speed and keep watch on the distance between it and the vehicle in front of it.

There’s a common misconception that driverless cars and driverless trucks rely on lane markers alone to make sense of the path before them. While this used to be the case for very early self-driving cars, our autonomous abilities have advanced to grand new heights. Today, lane markers (and the cameras that “see” them) are just one piece of the puzzle.

Self-driving vehicles also utilize radar, which is sometimes found in Lane Keep Assist programs that we have in our day-to-day cars. Radar adds an additional safety component to lane navigation. Using radio waves to detect objects like other cars and traffic cones, radar helps paint the picture of what the driving environment looks like. In the same vein, mapping allows a self-driving vehicle to utilize historical information about the road to navigate in conjunction with the other tools in its toolbox. These two tools work with lane marker detection to assess the environment, calculate the safest possible behavior, and execute that behavior.

Lane Keeping and Robotic Trucks

As proponents of safe and sustainable self-driving practices, our autonomous driving system keeps in line with regulatory and industry best practices throughout all lane keeping behaviors. Aside from its impact on safety, proper lane keeping allows us to operate predictably to drivers on the road around us, prove our product’s viability, and promote a safe self-driving future.

As we forge ahead with our driverless trucking development, we will continue to innovate, collaborate, and lead the way in advancing our autonomous driving system. Through ongoing research, development, and collaboration with our stakeholders and partners, we will further enhance our lane-keeping capabilities to meet the evolving needs and expectations of the industry and the public. Together, we’re driving the future of freight.

 

 

Are Torc’s driverless trucks currently on the road today?

Vehicle testing is paramount to our success. Our routes include public roads in Virginia and New Mexico, chosen in part for their diverse terrain, traffic, and weather conditions. Our testing on closed-course tracks includes Virginia Tech Transportation Institute and a Daimler Trucks facility in Madras, Oregon. In addition to test tracks, Torc has tested numerous vehicle platforms on public roads in 20 states over the past 14 years.