Search for: All records

We are experiencing a revolution in vehicle operation, with fully automated robotaxis deployed and available for public use in major U.S. markets in 2023. These vehicles, while imperfect, already are arguably safer than the average human driver. Despite this rapid progress, there remain significant research and development problems that must be addressed; beyond this, there is an underdeveloped workforce for skilled researchers, developers, and practitioners in these areas, a fact that may delay necessary advances. We have created and run for two years a National Science Foundation funded Research Experience for Undergraduates (NSF REU) focused on solving both unmet research needs, and workforce development and pipeline programs. In our REU, which makes use of simulation and two full-scale, street-legal drive-by-wire electric vehicles with perception, planning, and control capabilities, our primary goals include to (1) provide hands-on experiences to undergraduate students who otherwise might not have research opportunities to learn fundamental theories in autonomous vehicle development, (2) allow students to design algorithms to practice software development and evaluation using real vehicles on real test courses, (3) strengthen their confidence, self-guided capabilities, and research skills, and (4) increase the number of students, including those from diverse backgrounds and technical disciplines, interested in graduate programs to ultimately provide a quality research and development workforce to both academia and industry. Over the initial two years, a cohort of 8 diverse students each year learned fundamental self-driving and computer networking skills including coding for drive-by-wire vehicles, computer vision, use of localization, and interpretation of richer sensor data, as well as network and communication protocols. The students were introduced to research ideation and publishing concepts, mentored in designing and testing hypotheses, and then involved in two challenges related to self-driving and networked vehicles. Two teams of 4 designed, implemented, tested various self-drive and V2X algorithms using real vehicles on a test course, analyzed/evaluated test results, wrote technical reports, and delivered presentations. After the summer program was over, the technical reports were published in peer reviewed conferences and journals. Survey results show that students attained significant & real-world computer science skills in autonomous vehicle development leveraging real vehicles available. The programs also increased research career interests and strengthened students’ confidence, self-guided capabilities, and research skills, while additionally supporting the development of workshop materials, simulators, and related content that provide valuable resources for others planning to develop an undergraduate curriculum to teach self-drive and networked vehicle development. 

This research develops, compares, and analyzes both a traditional algorithm using computer vision and a deep learning model to deal with dynamic road conditions. In the final testing, the deep learning model completed the target of five laps for both the inner and outer lane, whereas the computer vision algorithm only completed almost three laps for the inner lane and slightly over four laps for the outer. After conducting statistical analysis on the results of our deep learning model by finding the p-value between the absolute error and squared error of the self-driving algorithm in the outer lane and inner lane, we find that our results are statistically significant based on a two-tailed T test with unequal variances where the p-value for absolute error is 0.009, and 0.001 for squared error. Self-driving vehicles are not only complex, but they are growing in necessity—therefore, finding an optimal solution for lane detection in dynamic conditions is crucial to continue innovation. 

Self-driving and automated vehicles rely on a comprehensive understanding of their surroundings and one another to operate effectively. While the use of sensors may allow the vehicles to directly perceive their environments, there are instances where information remains hidden from a vehicle. To address this, vehicles can transmit information between each other, enabling over-the-horizon awareness. We create a Robot Operating System simulation of vehicle-to-everything communication. Then, using two real-life electric vehicles equipped with global positioning systems and cameras, we aggregate time, position, and navigation information into a central database on a roadside unit. Our model uses an image classification deep learning model to detect obstacles on the road. Next, we create a web-based graphical user interface that automatically updates to display the vehicles and obstacles from the database. Finally, we use an occupancy grid to predict vehicle trajectories and prevent potential collisions. Our deep learning model has a precision-recall score of 0.995 and our system works across many devices. In the future, we aim to recognize a broader range of objects, including pedestrians, and use multiple roadside units to widen the scope of the model. 

Robust lane-following algorithms are one of the main challenges in developing effective automated vehicles. In this work, a team of four undergraduate students designed and evaluated several automated lane-following algorithms using computer vision as part of a Research Experience for Undergraduates program funded by the National Science Foundation. The developed algorithms use the Robot Operating System (ROS) and the OpenCV library in Python to detect lanes and implement the lane-following logic on the road. The algorithms were tested on a real-world test course using a street-legal vehicle with a high-definition camera as input and a drive-by-wire system for output. Driving data were recorded to compare the performance of human driving to that of the self-driving algorithms on the basis of three criteria: lap completion time, lane positioning infractions, and speed limit infractions. The evaluation of the data showed that the human drivers successfully completed every lap with zero infractions at a 100% success rate in varied weather conditions, whereas, our most reliable algorithms had a success rate of at least 70% with some lane positioning infractions and at lower speeds. 

Reliable lane-following is one of the most important tasks for an automated vehicle or ADAS. The intent of this project was to design and evaluate multiple lane-following algorithms for an automated vehicle using computer vision. The implemented algorithms' performance was then evaluated on a testing course and compared with a human driver. ROS and OpenCV were used to detect and follow lanes on the road. A street-legal vehicle with a high-definition camera and drive-by-wire system was used to implement and evaluate driving data. Each algorithm was evaluated based on time for completion, speed limit infractions, and lane positioning infractions. The recorded evaluation data determined the most reliable lane-following algorithm. All of our algorithms had a success rate of at least 60% on certain lanes of the testing course.