Christopher Goodin

Artificial intelligence (AI) in autonomous vehicles (AVs) is gaining much focus on safety, comfort, and efficiency. The goal is to provide the driver with assistance and mimic human driving. Deep reinforcement learning (DRL), specifically the deep deterministic policy gradient (DDPG) is often used for longitudinal velocity control, utilizing the velocities of the vehicles and the distance between them. DDPG has been found effective in many studies when it is embedded with a self-supervised learning (SSL) method. In this paper, a framework for longitudinal velocity control is proposed using SSL and DDPG frameworks. The inputs of the DDPG networks have been replaced by the outputs of the SSL network. The primary objective of this SSL is to enable the model to accurately predict future states based on current states and actions. The input features of the datasets have been modified so that the DDPG model can get additional information that helps the model make better predictions. These features include the distance between vehicles, ego vehicle velocity, acceleration, jerk, and lead vehicle relative velocity, and estimated velocity. Furthermore, a custom reward function is designed to account for safety, driving comfort, negative impact, driving aggressiveness, and fuel efficiency. In order to evaluate the model, the algorithm has been trained and tested on a variety of datasets, including simulated and real-world data. The analysis demonstrates that the new architecture maintains strong robustness across various datasets and outperforms the current state-of-the-art models.

Autonomous ground vehicles (AGVs) operating in off-road terrain are influenced by a variety of factors that are unique to the off-road environment, especially the presence of vegetation. Accurately simulating the performance of AGV requires the creation of off-road virtual worlds that realistically present the characteristics of vegetation to simulations of the tires, chassis, and sensor systems. In this work, we present the development and implementation of a coupled soil moisture and vegetation growth model for generating synthetic off-road terrains for use in AGV simulations. These digital scenes have high geometric fidelity for simulating sensor systems used on AGV like terrestrial lidar. The vegetation model uses stored carbohydrates to predict growth cycles and takes multiple phenomenon into account including soil moisture, weather conditions, and seasonal variations. Results of AGV simulations in synthetic terrains are presented as a demonstration of the models utility.

The perception of vegetation is a critical aspect of off-road autonomous navigation, and consequentially a critical aspect of the simulation of autonomous ground vehicles (AGVs). Representing vegetation with triangular meshes requires detailed geometric modeling that captures the intricacies of small branches and leaves. In this work, we propose to answer the question, “What degree of geometric fidelity is required to realistically simulate lidar in AGV simulations?” To answer this question, in this work we present an analysis that determines the required geometric fidelity of digital scenes and assets used in the simulation of AGVs. Focusing on vegetation, we use a comparison of the real and simulated perceived distribution of leaf orientation angles in lidar point clouds to determine the number of triangles required to reliably reproduce realistic results. By comparing real lidar scans of vegetation to simulated lidar scans of vegetation with a variety of geometric fidelities, we find that digital tree models (meshes) need to have a minimum triangle density of >1600 triangles per cubic meter in order to accurately reproduce the geometric properties of lidar scans of real vegetation, with a recommended triangle density of 11,000 triangles per cubic meter for best performance. Furthermore, by comparing these experiments to past work investigating the same question for cameras, we develop a general “rule-of-thumb” for vegetation mesh fidelity in AGV sensor simulation.

The transportation system is moving towards higher levels of autonomy. Longitudinal speed control, or adaptive cruise control (ACC), is intended to provide driver assistance by controlling vehicle speed and maintaining a safe distance from the preceding vehicle. Artificial intelligence (AI) and machine learning (ML) paved the way for robust navigation and decision-making in complex environments. In this paper, we propose a comprehensive framework for speed control using a deep reinforcement learning algorithm for safety and comfort. We incorporate the deep deterministic policy gradient (DDPG) framework with an attention mechanism for this. Many works achieved smooth vehicle control output from the DDPG algorithm with different network structures, but we demonstrate that adding an attention mechanism improves the overall performance. The baseline DDPG framework is based on fully connected layers. However, when we bring the attention mechanism to the DDPG model, it helps to increase focus on the most important features and enhances the overall model effectiveness. We also designed a custom reward function as our priority is to improve overall safety and comfort. The efficacy of our model is evaluated in comparison to the baseline DDPG model, emphasizing the influence of the attention mechanism. We have performed an ablation study to determine the impact of the number of layers and neurons in the hidden layer. In addition, we tested this model on diverse datasets (simulated and real), including some unknown scenarios in the testing datasets. We demonstrate that our architecture has outperformed the state-of-the-art model, and is quite robust across different datasets.

This field report presents the results of a study of obstacle detection and avoidance (ODOA) by an autonomous ground vehicle (AGV) in off‐road driving conditions. This study included both real and simulated testing of the AGV and served as the third and final phase of a 3‐year research project studying the influence of environmental conditions over autonomous driving. We compare and contrast the results of the real and experimental field testing and report our findings on the influence of soft soil in ODOA performance by an AGV. We find that rutting in soft soil results in higher throttle effort but lower steering effort and speed. The results of these experiments will inform future studies of autonomous driving on off‐road environments.

DOI: https://doi.org/10.1103/PhysRevC.107.039905

Machine-learning (ML) requires human-labeled “truth” data to train and test. Acquiring and labeling this data can often be the most time-consuming and expensive part of developing trained models of convolutional neural networks (CNN). In this work, we show that an automated workflow using automatically labeled synthetic data can be used to drastically reduce the time and effort required to train a machine learning algorithm for detecting buildings in aerial imagery acquired with low-flying unmanned aerial vehicles. The MSU Autonomous Vehicle Simulator (MAVS) was used in this work, and the process for integrating MAVS into an automated workflow is presented in this work, along with results for building detection with real and simulated images.

Autonomous ground vehicles (AGVs) operating collaboratively have several advantages over vehicles operating alone. An AGV team may be more resilient and efficient than a single AGV. Other advantages of AGV teams include increased coverage and multiple viewing angles of terrain features as well as resistance to failure from any single AGV. Additionally, AGV teams can explore large terrains more quickly and thoroughly than a single system. In this work, the feasibility of using a team of high-mobility AGV to explore a navigation corridor, map the terrain, and autonomously flag obstacles for future navigation is evaluated. Focusing on negative obstacles, the value of using multiple vehicles to map a navigation corridor is quantified. This study is the first to evaluate large teams of AGV collaborating in realistic off-road, 3D environments. The feasibility of the large-scale AGV team is demonstrated while avoiding the high cost of purchasing and testing large numbers of vehicles using the Mississippi State University autonomous vehicle simulator (MAVS), a high-fidelity, physics-based simulation tool. The cost and benefits of increasing the AGV team size are evaluated. The simulation results show how factors like fuel use, map coverage, and obstacle detection are influenced by increasing numbers of AGV in the team. The simulation architecture is presented and experiments quantifying the performance of the simulator are shown. Finally, a model for evaluating the tradeoff between mission effectiveness and fuel use is developed and presented to demonstrate the utility of this approach.

Project 1.31 Evaluating Sensitivity of Autonomous Algorithms to Sensor Error and Environmental Conditions While much effort has been devoted to the development of autonomous navigation algorithms in recent years, techniques and procedures for quantifying the performance of these algorithms have been developed on an ad-hoc, case-by-case basis. In the first year of this work, we