Image quality analysis (IQA) techniques proposed as a novel method to evaluate digital-twin environment fidelity and predict the simulation-to-real-world (sim2real) transferability of detection models by comparing subsets of synthetic and real-world image pair data. IQA can quantify similarity and explain modeled feature limitations that correlated to an expected potential of sim2real transfer performance for detection models, providing an initial assessment of validation for virtual autonomy results. Of the metrics evaluated, the structural similarity (SSIM) and complex wavelet SSIM (CW-SSIM) indexes identified apparent trends of increasing potential for sim2real success with an increase in the corresponding IQA metric. SSIM and CW-SSIM indexes respectively indicated a significant increase in potential of detection model accuracies at approximately 0.10 and 0.15. The SSIM-based indexes gave insight to image pair contrast, structure, and luminance feature similarities to describe the digital-twin environment fidelity. An open-source image quality analysis calculator application, Synthetic Image Quality Analysis Calculator (SIQAC), was developed to automate IQA evaluations of image pairs and predict the potential of sim2real transferability for both classifiers and object detectors. The work was based on human vision system and compression algorithm studies.

In Indiana, roadside mowing operations span over 11,000 miles and are performed by contracted crews using tractors and hand-held trimmers, exposing workers to significant safety risks, including reported casualties. Autonomous mowers offer a promising solution to enhance safety, but large-scale deployment requires rigorous testing to ensure reliability. This study addresses this need by developing digital-twin roadside environments to evaluate autonomous mowing systems. The method generates realistic roadways that accurately replicate real-world conditions, achieving single-float rounding planar accuracy over tens of meters, enabling risk-free and comprehensive testing. Leveraging remote sensing data, model libraries, and a georeferenced data mapping tool, the approach overcomes challenges in scalability and fidelity, even in locations with low-density source data. Digital roadway construction accelerates testing, enhances the realism of virtual deployments, and supports the development of autonomous mowing technologies. Furthermore, this approach has broader applications for generating test environments for various roadway-related systems.

Roadside mowing is a necessary, high-risk occupation to ensure visibility for on-road traffic. In Indiana, contractors are hired by the Indiana Department of Transportation (INDOT) to mow over 11,000 miles of roadsides using tractors with flex-wing mowers and hand-held string trimmers. Operations have significant risks to all crew members. Autonomous mowers have gained interest as a potential solution to reduce the risk. INDOT has expressed interest in how to evaluate the technological readiness of autonomous mowers. This work developed ROWMow Sim, a digital-twin autonomous mowing simulator that can test and evaluate roadside mowing operations. ROWMow Sim generates virtual environments of real-world roadways, providing an informative testbed with zero risk. Virtual models of sensors and vehicles were developed to be tested in virtual roadways prior to real-world deployment. ROWMow Sim provides rapid creation of various environment conditions, repeatable tests, ground truth data analysis, mower configuration studies, and direct simulation-to-real-world transfers. Autonomous navigation strategies developed on a virtual string trimming vehicle were transferred directly to the physical vehicle. ROWMow Sim contributes to the exploration of safer and more efficient vegetation management operations for INDOT and the on-road traffic.

Maize research has grown an interest in exploring the capability to produce a uniform orientation of maize leaves in a field to potentially improve the sunlight capture and increase yield. Previous research tracked leaf orientations to further studies of exploring the potential of a relationship between seed and leaf orientations. Tracking ear orientation may offer a higher throughput insight and the ear tends to follow the leaf orientation. We proposed vision-based machine learning (ML) models to detect ears and estimate the orientation passively at a high rate. To evaluate the proposed method as a proof-of-concept, we had to overcome both seasonal limitations for real-world data generation and manual labor hours for producing a large dataset. The solution was the development of a synthetic maize field and an automated synthetic data generation and labeling pipeline, enabling year-round maize orientation research. The maize field three-dimensional scanning method circumvented millimeter-thin structure challenges to generate low-fidelity leaf structures at scale. The synthetic data pipeline was scaled-up from simpler proof-of-concept objects to the complex maize ear. The synthetic pipeline determined the ear angle could be passively estimated in real-time. The synthetic maize ML models were able to identify the best to worst performing angle estimation model types for the real world. Synthetic ear object detectors were unable to transfer between synthetic and real-world domains. A ResNet18 classifier trained on synthetic ear data tested at 66% accuracy in the real world.