Search Results

Autonomous vehicle operation is dependent upon accurate position estimation and thus a major concern of implementing the autonomous navigation is obtaining robust and accurate data from sensors. This is especially true, in case of Inertial Measurement Unit (IMU) sensor data. The IMU consists of a 3-axis gyro, 3-axis accelerometer, and 3-axis magnetometer. The IMU provides vehicle orientation in 3D space in terms of yaw, roll and pitch. Out of which, yaw is a major parameter to control the ground vehicle’s lateral position during navigation. The accelerometer is responsible for attitude (roll-pitch) estimates and magnetometer is responsible for yaw estimates. However, the magnetometer is prone to environmental magnetic disturbances which induce errors in the measurement.

Autonomous ground vehicles can use a variety of techniques to navigate the environment and deduce their motion and location from sensory inputs. Visual Odometry can provide a means for an autonomous vehicle to gain orientation and position information from camera images recording frames as the vehicle moves. This is especially useful when global positioning system (GPS) information is unavailable, or wheel encoder measurements are unreliable. Feature-based visual odometry algorithms extract corner points from image frames, thus detecting patterns of feature point movement over time. From this information, it is possible to estimate the camera, i.e., the vehicle’s motion. Visual odometry has its own set of challenges, such as detecting an insufficient number of points, poor camera setup, and fast passing objects interrupting the scene. This paper investigates the effects of various disturbances on visual odometry.

Over the last decade, tremendous amount of research and progress has been made towards developing smart technologies for autonomous vehicles such as adaptive cruise control, lane keeping assist, lane following algorithms, and decision-making algorithms. One of the fundamental objectives for the development of such technologies is to enable autonomous vehicles with the capability to avoid obstacles and maintain safety. Automobiles are real-world dynamical systems - possessing inertia, operating at varying speeds, with finite accelerations/decelerations during operations. Deployment of autonomy in vehicles increases in complexity multi-fold especially when high DOF vehicle models need to be considered for robust control. Model Predictive Control (MPC) is a powerful tool that is used extensively to control the behavior of complex, dynamic systems. As a model-based approach, the fidelity of the model and selection of model-parameters plays a role in ultimate performance.

Simultaneous Localization and Mapping (SLAM) algorithms are extensively utilized within the field of autonomous navigation. In particular, numerous open-source Robot Operating System (ROS) based SLAM solutions, such as Gmapping, Hector, Cartographer etc., have simplified deployments in application. However, establishing the accuracy and precision of these ‘out-of-the-box’ SLAM algorithms is necessary for improving the accuracy and precision of further applications such as planning, navigation, controls. Existing benchmarking literature largely focused on validating SLAM algorithms based upon the quality of the generated maps. In this paper, however, we focus on examining the localization accuracy of existing 2-dimensional LiDAR based indoor SLAM algorithms. The fidelity of these implementations is compared against the OptiTrack motion capture system which is capable of tracking moving objects at sub-millimeter level precision.

In the next 20 years fully autonomous vehicles are expected to be in the market. The advance on their development is creating paradigm shifts on different automotive related research areas. Vehicle interiors design and human vehicle interaction are evolving to enable interaction flexibility inside the cars. However, most of today’s vehicle manufacturers’ autonomous car concepts maintain the steering wheel as a control element. While this approach allows the driver to take over the vehicle route if needed, it causes a constraint in the previously mentioned interaction flexibility. Other approaches, such as the one proposed by Google, enable interaction flexibility by removing the steering wheel and accelerator and brake pedals. However, this prevents the users to take control over the vehicle route if needed, not allowing them to make on-route spontaneous decisions, such as stopping at a specific point of interest.

This paper introduces an eco-driving highway cruising algorithm based on optimal control theory that is applied to a conventionally-powered connected and automated vehicle. Thanks to connectivity to the cloud and/or to infrastructure, speed limit and slope along the future route can be known with accuracy. This can in turn be used to compute the control variable trajectory that will minimize energy consumption without significantly impacting travel time. Automated driving is necessary to the implementation of this concept, because the chosen control variables (e.g., torque and gear) impact vehicle speed. An optimal control problem is built up where quadratic models are used for the powertrain. The optimization is solved by applying Pontryagin’s minimum principle, which reduces the problem to the minimization of a cost function with parameters called co-states.

Emerging autonomous driving technologies, with emergency navigating capabilities, necessitates innovative vehicle steering methods for operators during unanticipated scenarios. A reconfigurable “plug and play” steering system paradigm enables lateral control from any seating position in the vehicle’s interior. When required, drivers may access a stowed steering input device, establish communications with the vehicle steering subsystem, and provide direct wheel commands. Accordingly, the provision of haptic steering cues and lane keeping assistance to navigate roadways will be helpful. In this study, various steering devices have been investigated which offer reconfigurability and haptic feedback to create a flexible driving environment. A joystick and a robotic arm that offer multiple degrees of freedom were compared to a conventional steering wheel.

Robust autonomous navigation in unstructured environments is an unsolved problem and critical to the operation of autonomous military and rescue ground vehicles. Two-dimensional path planners operating on occupancy grids or costs maps can produce infeasible paths when the operational area includes complex terrain. Recently, sample-based path planners that plan on LiDAR-acquired point-cloud maps have been proposed. These approaches require no discretization of the operational area and provide direct pose estimation by modeling vehicle and terrain interaction. In this paper, we show that direct sample-based path planning on point clouds is effective and robust in unstructured environments. Robustness is demonstrated by completing a system parameter sensitivity analysis of the system in an Unreal simulation environment and partnered with field validation.

Off-road vehicle operation requires constant decision-making under great uncertainty. Such decisions are multi-faceted and range from acquisition decisions to operational decisions. A major input to these decisions is terrain information in the form of soil properties. This information needs to be propagated to path planning algorithms that augment them with other inputs such as visual terrain assessment and other sensors. In this sequence of steps, many resources are needed, and it is not often clear how best to utilize them. We present an integrated approach where a mission’s overall performance is measured using a multiattribute utility function. This framework allows us to evaluate the value of acquiring terrain information and then its use in path planning. The computational effort of optimizing the vehicle path is also considered and optimized. We present our approach using the data acquired from the Keweenaw Research Center terrains and present some results.

Testing any new safety technology of Autonomous Vehicles (AV) and Advanced Driver Assistance Systems (ADAS) requires simulation-based validation and verification. The specific scenarios used for testing, outline incidences of accidents or near-miss events. In order to simulate these scenarios, specific values for all the above parameters are required including the ego vehicle model. The ‘criticality’ of a scenario is defined in terms of the difficulty level of the safety maneuver. A scenario could be over-critical, critical, or under-critical. In over-critical scenarios, it is impossible to avoid a crash whereas, for under-critical scenarios, no action may be required to avoid a crash. The criticality of the scenario depends on various parameters e.g. speeds, distances, road/tire parameters, etc. In this paper, we propose a definition of criticality metric and identify the parameters such that a scenario becomes critical.

In this paper, a reinforcement learning-based method is proposed to adapt autonomous vehicle passengers’ expectation of comfort through in-situ human-vehicle interaction. Ride comfort has a significant influence on the user’s experience and thus acceptance of autonomous vehicles. There is plenty of research about the motion planning and control of autonomous vehicles. However, limited studies have explicitly considered the comfort of passengers in autonomous vehicles. This paper studies the comfort of humans in autonomous vehicles longitudinal autonomous driving. The paper models and then improves passengers’ feelings about autonomous driving behaviors. This proposed approach builds a control and adaptation strategy based on reinforcement learning using human’s in-situ feedback on autonomous driving. It also proposes an adaptation of humans to autonomous vehicles to account for improper human driving expectations.

Integrated modeling of vehicle, tire and terrain is a fundamental challenge to be addressed for off-road autonomous navigation. The complexities arise due to lack of tools and techniques to predict the continuously varying terrain and environmental conditions and the resultant non-linearities. The solution to this challenge can now be found in the plethora of data driven modeling and control techniques that have gained traction in the last decade. Data driven modeling and control techniques rely on the system’s repeated interaction with the environment to generate a lot of data and then use a function approximator to fit a model for the physical system with the data. Getting good quality and quantity of data may involve extensive experimentation with the physical system impacting developer’s resource. The process is computationally expensive, and the overhead time required is high.

Access control enforces security policies for controlling critical resources. For V2X (Vehicle to Everything) autonomous military vehicle fleets, network middleware systems such as ROS (Robotic Operating System) expose system resources through networked publisher/subscriber and client/server paradigms. Without proper access control, these systems are vulnerable to attacks from compromised network nodes, which may perform data poisoning attacks, flood packets on a network, or attempt to gain lateral control of other resources. Access control for robotic middleware systems has been investigated in both ROS1 and ROS2. Still, these implementations do not have mechanisms for evaluating a policy's consistency and completeness or writing expressive policies for distributed fleets. We explore an RBAC (Role-Based Access Control) mechanism layered onto ROS environments that uses local permission caches with precomputed truth tables for fast policy evaluation.

The software architecture behind modern autonomous vehicles (AV) is becoming more complex steadily. Safety verification is now an imminent task prior to the large-scale deployment of such convoluted models. For safety-critical tasks in navigation, it becomes imperative to perform a verification procedure on the trajectories proposed by the planning algorithm prior to deployment. Signal Temporal Logic (STL) constraints can dictate the safety requirements for an AV. A combination of STL constraints is called a specification. A key difference between STL and other logic constraints is that STL allows us to work on continuous signals. We verify the satisfaction of the STL specifications by calculating the robustness value for each signal within the specification. Higher robustness values indicate a safer system. Model Predictive Control (MPC) is one of the most widely used methods to control the navigation of an AV, with an underlying set of state and input constraints.

Image segmentation has historically been a technique for analyzing terrain for military autonomous vehicles. One of the weaknesses of image segmentation from camera data is that it lacks depth information, and it can be affected by environment lighting. Light detection and ranging (LiDAR) is an emerging technology in image segmentation that is able to estimate distances to the objects it detects. One advantage of LiDAR is the ability to gather accurate distances regardless of day, night, shadows, or glare. This study examines LiDAR and camera image segmentation fusion to improve an advanced driver-assistance systems (ADAS) algorithm for off-road autonomous military vehicles. The volume of points generated by LiDAR provides the vehicle with distance and spatial data surrounding the vehicle.

Autonomous vehicles (AV) are sophisticated systems comprising various sensors, powerful processors, and complex data processing algorithms that navigate autonomously to their respective goals. Out of several functions performed by an AV, one of the most important is developing situational intelligence to predict collision-free future trajectories. As an AV operates in environments consisting of various entities, such as other AVs, human-driven vehicles, and static obstacles, developing situational intelligence will require a collaborative approach. The recent developments in artificial intelligence (AI) and deep learning (DL) relating to AVs have shown that DL-based models can take advantage of information sharing and collaboration to develop such intelligence.

In the United States and worldwide, 38,824 and 1.35 million people were killed in vehicle crashes during 2020. These statistics are tragic and indicative of an on-going public health crisis centered on automobiles and other ground transportation solutions. Although the long-term US vehicle fatality rate is slowly declining, it continues to be elevated compared to European countries. The introduction of vehicle safety systems and re-designed roadways has improved survivability and driving environment, but driver behavior has not been fully addressed. A non-confrontational approach is the evaluation of driver behavior using onboard sensors and computer algorithms to determine the vehicle’s “mistrust” level of the given operator and the safety of the individual operating the vehicle. This is an inversion of the classic human-machine trust paradigm in which the human evaluates whether the machine can safely operate in an automated fashion.

This paper describes a multi-sensor data set, suitable for testing algorithms to detect and track pedestrians and cyclists, with an autonomous vehicle’s sensor suite. The data set can be used to evaluate the benefit of fused sensing algorithms, and provides ground truth trajectories of pedestrians, cyclists, and other vehicles for objective evaluation of track accuracy. One of the principal bottlenecks for sensing and perception algorithm development is the ability to evaluate tracking algorithms against ground truth data. By ground truth we mean independent knowledge of the position, size, speed, heading, and class of objects of interest in complex operational environments. Our goal was to execute a data collection campaign at an urban test track in which trajectories of moving objects of interest are measured with auxiliary instrumentation, in conjunction with several autonomous vehicles (AV) with a full sensor suite of radar, lidar, and cameras.

Autonomous driving technologies can provide better safety, comfort and efficiency for future transportation systems. Most research in this area has mainly been focused on developing sensing and control approaches to achieve various autonomous driving functions. Very little of this research, however, has studied how to efficiently handle sensing exceptions. A simple exception measured by any of the sensors may lead to failures in autonomous driving functions. The autonomous vehicles are then supposed to be sent back to manufacturers for repair, which takes both time and money. This paper introduces an efficient approach to make human drivers able to online teach autonomous vehicles to drive under sensing exceptions. A human-vehicle teaching-and-learning framework for autonomous driving is proposed and the human teaching and vehicle learning processes for handling sensing exceptions in autonomous vehicles are designed in detail.

Learning only from a small set of examples remains a huge challenge in machine learning. Despite recent breakthroughs in the applications of neural networks, the applicability of these techniques has been limited by the requirement for large amounts of training data. What’s more, the standard supervised machine learning method does not provide a satisfactory solution for learning new concepts from little data. However, the ability to learn enough information from few samples has been demonstrated in humans. This suggests that humans may make use of prior knowledge of a previously learned model when learning new ones on a small amount of training examples. In the area of autonomous driving, the model learns to drive the vehicle with training data from humans, and most machine learning based control algorithms require training on very large datasets. Collecting and constructing training data set takes a huge amount of time and needs specific knowledge to gather relevant information.