Search

Traditional tractor ground leveling operation applies a manual process with no electronic assistance. Automated Ground Leveling (AGL) will increase quality of leveling and operator comfort. This paper outlines a machine learning approach using Artificial Neural Network (ANN). The proposed AGL uses tractor inclined angle and leveling error as inputs. The target output is the tractor scraper implement raise or lower command. The equations to run simulations are formulated and applied to the model and verified during simulations. The details can be found in section IV of this paper. John Deere StarFire 6000 GPS receiver is proposed to be the device to obtain latitude, longitude, altitude and an IMU device to obtain pitch/angling data of tractor. The proposed inputs and target output proved to be effective in producing a set of weights and biases that learns to control the scraper implement. Twenty (20) ANN trainings were conducted using the same set of training data. Out of the twenty trainings, three sets of trained weights and biases outperformed the training set. The best trained weights and biases produced an RMS error of 0.50449 compared to human training data RMS error of 0.593, which was about 14.9% improvement. The algorithm recognizes the goal of staying close to the ground reference line. This paper provides a brief review on ANN for clarity and applies it to the AGL.

A novel approach for predicting collision warning levels using a classifier that interprets a window of present and past relative positions between the ego vehicle and front object is considered. The premise assumes that the ego vehicle is equipped with a LiDAR, radar or camera that measures the relative distance. A machine learning approach is presented where the feature is the time-to-collision (TTC) which can be determined from Kalman filter estimate of distance, speed and acceleration. The classifier produces classes of warning labels that combines alert levels with watch levels. First part of the classifier called predictor classifier 1, produces the alert levels by evaluating the TTC value of current instance; however, the second part of the classifier called predictor classifier 2, produces the watch labels by applying a novel mathematical algorithm that interprets past values of TTC in a given window. Alert and watch labels are merged together by warning labels generator to produce the final labels for the dataset. This paper presents the formulation and simulation results for the machine learning approach that utilizes neural networks. A future extension of the result will address a deep learning approach where the Kalman filtering & TTC features will be eliminated.

An agile process is a frequent redesign technique that is suitable for quick adaptation of various tasks, in contrast to a more time-consuming high-level design. In this paper, we describe the concept of an agile development process to convert a by-wire electric vehicle to meet Intelligent Ground Vehicle Competition (IGVC) Self-Drive Challenges. We converted a commercial electric vehicle into a basic self-driving by-wire electric vehicle within a limited development time to satisfy the rule. A key factor for enabling agile development is the Robot Operating Systems (ROS) environment for integrating multiple microcontrollers with sensor and actuator modules. The approach allows the rapid installation of hardware components implemented as nodes, and rapid debugging of software programs implemented as topics. Employment of ROS allows frequent reassessments that enable rapid troubleshooting and debugging. This paper illustrates how a team of eight students with three advisors were able to successfully integrate hardware and software to meet specific autonomous driving challenges in a matter of one and a half months. The advantages of the agile development process were confirmed in the team meeting the IGVC self-driving challenges.

In this paper, a systematic approach is presented and a physical experiment substantiated for a SISO wind tunnel adaptive control system. This approach includes a two-step adaptation scheme: The first comprises a triggered system identification process and the second an adapting controller system. To verify this integrated systems engineering approach, a physical wind tunnel was constructed with an axial fan and airspeed sensors. A real-time adapting system model, coupled with an adaptive control system is interfaced with the wind tunnel. The real-time data from the system in operation is analyzed and used to influence how the controller adapts through system identification techniques as well as PID tuning criteria.

The well-known adaptive neuro-fuzzy inference system (ANFIS) uses a combination of least-square estimation (LSE) and gradient descent back-propagation methods to model a training data set. In this paper, we show that the rate of convergence of ANFIS can be very much improved by using a combination of LSE and Levenberg-Marquardt algorithm (LMA). The improved ANFIS converges more closely and significantly more rapidly to the data. Detail explanation of the proposed ANFIS is presented, and its validity is verified via simulation.

Advanced driver's assist systems for optional self-driving vehicles require highly reliable navigation capabilities to ensure dependable autonomous function of vehicle safety and driver assistance. Radars, lidars and cameras have found widespread application in nearly all vehicle brands for this purpose. In this paper we present the implementation of these sensors for constructing a 360-degree field of view of the vehicle surrounding. A stereo vision system is complemented by several long, mid and short-range radar sensors. The data gathered from the advanced sensors is prioritized and fused to create a model of the surrounding. The system is being tested on an autonomous car vehicle. In particular, this paper discusses three clustering algorithms that were tested.

Flocking is a term that describes the behavior of a group of birds (a flock) in flight, or the swarming behavior of insects. This paper presents detailed information about how to use the flocking techniques to control a group of embedded controlled systems “Boids” such as ground systems (robotic vehicles/ swarm robots) in 2D and/or aerial systems (quadrotors) in 3D. Each one of these systems moves inside/outside of a building collectively to reach a target. This paper presents flocking behavior implemented on a server based control, which will process each of the boids properties e.g. position, speed & target. Subsequently the server will assign the appropriatee move to a specific boid. The calculated information will be used locally to control and direct the movements/flocking for each boid in the group. A simulation technique and detailed flow chart is presented. In addition to the famous Reynolds' Rules for flocking two other rules are presented; one for targeting and another for avoiding obstacles. The targeting rule was implemented to ensure that the flock moves collectively towards a common target; it was implemented by creating a target velocity parameter. The avoiding obstacles rule was implemented to ensure that the flock doesnt collide with obstacles that they could be in the path; it was implemented by creating an avoiding velocity parameter.

This paper presents a novel approach to control the formation of a group of unmanned ground vehicles in an outdoor environment. Using Lyapunov stability analysis, a closed loop control law for a differential-drive vehicle is derived to maintain its target position in the formation. This control law is combined with an optimal control strategy to avoid obstacles. A simple fuzzy logic supervisor balances the weight each algorithm has on the output control signals by gradually allowing the obstacle avoidance to take over the steering as obstacles become close. Simulations have shown this supervised optimal control strategy to be an effective algorithm that seamlessly allows the group of vehicles to temporarily break formation to avoid obstacles.

In this paper, a technique for designing a Ball and Beam System controller based on a nonlinear backstepping design is presented. A control law is developed, and asymptotic stability based on a Lyapunov stability criterion is satisfied. The goal is to design a controller for the ball position with the least overall energy consumption and minimum overshoot. In order to illustrate the efficiency of the proposed control strategy, the simulations are demonstrated and compared with [1]. The results show that the nonlinear backstepping design gives a smoother performance and needs less input magnitude compared to the LQR design. Moreover, the robustness of the proposed design with respect to parameter variations and different reference inputs is examined, and the simulations validated the control scheme.

In this paper, an optimal control approach is used to solve a two-dimensional path planning problem for a differential drive mobile robot. Two optimal control algorithms are presented and analyzed, which consist of a novel implementation of a linear quadratic tracker (LQT), and a dynamic programming (DP) scheme. The algorithms are applied to the task of GPS navigation with obstacle avoidance. The methods aim to find an optimal path where the tracking error to the GPS target is minimized, while avoiding the obstacles present on the vehicle's map. The LQT algorithm minimizes the tracking error to the goal point, while simultaneously maximizing the distance to obstacles. It also makes use of a fuzzy logic system to adjust the optimization parameters according to different environmental scenarios.

A local positioning system (LPS) for tracking mobile robots using a newly developed ultra-wideband (UWB) ranging radio technology is presented in this paper. However, measured ranges from these radios often have uncertain biases and large sporadic errors due to multipath and attenuation effect. A fuzzy neighborhood tracking filtering technique was developed to deal with the range outlier problems. A progressive update trilateration filter technique is introduced. The paper then describes a UWB LPS based on triangulation or trilateration of the ranges, which is fused with other vehicle kinematics/dynamics sensors including a compass, rate gyros and wheel speed sensors. The UWB LPS was applied to navigation and guidance of an experimental autonomous mobile robot. Simulation theoretical and real-time experimental results were validated against the true results of the experiment recorded by video.

This paper addresses challenges in managing complex automotive powertrain control systems by employing a heuristic method to determine complex setpoints. Two main areas of focus are the fuzzy determination of the driver characteristics and the gear selection for discrete automatic transmissions. The driver characterization determines how aggressive or docile a driver is and has a weighted impact on the gear selection. The gear selection process captures transmission performance fundamentals and infuses them with logic to eliminate undesired or physically impossible gears.

The tuning process of Fuzzy Logic Controllers (FLCs) through trial-and-error is usually a tedious and time-consuming task. The significant amount of computation imposed by current techniques used in self organized FLCs coupled with the need for a shortened development cycle in today's fast manufacturing environment, make existing methods practically obsolete. This paper introduces and demonstrates a methodology to simplify the tuning process of self organized FLC. Through identifying what will be described as core membership functions MFs, limiting the tuning process to core MFs, utilizing knowledge from domain experts, and mathematically derive other important MFs from core MFs once the tuning is completed, it will be demonstrated in this paper that the tuning process of FLCs can be greatly simplified.

Thousands of people die every year in car accidents. Current restraint systems such as conventional seatbelts and airbags are designed to accommodate several occupant/crash scenarios and lack the capability of varying their output during the impact event. This paper introduces and demonstrates the benefits of real time control of a novel adaptive restraint system. Through applying intelligent control techniques, it will be shown in this paper that occupant injuries during frontal crashes can be minimized to much lower levels than what can be achieved today with conventional restraint systems. Our control system will be designed and tested first using a lumped parameter model and second on a complete crash simulation environment.

In every vehicle collision, the work-energy principle must be satisfied, i.e., the work done in stopping the driver must be equal to the driver's kinetic energy. Existing restraint systems such as airbags and seatbelts are designed to minimize injuries by smoothly absorbing the kinetic energy of the occupant during a crash event. Advanced systems are being developed to tailor their level of restraining to the severity of the crash, weight, size and position of the occupant. Starting with a lumped parameter model and ending with a complete crash simulation environment, this paper introduce and demonstrate the benefits of real time control of a novel adaptive restraint system using intelligent control techniques.

Neural networks are known for their ability to learn and classify patterns based on certain criteria defined within the training process. Fuzzy ARTMAP neural networks are examples of such systems where the output is decided based on the input/output pattern training scheme. In this research, we build a fuzzy ARTMAP like neural network that depends on an adaptive Euclidian distance neighborhood rather than the fuzzy AND neighborhood in deciding the network output. It is a supervised input/output clustering algorithm that calculates the Euclidean distance between input patterns and system stored categories (neurons) to determine the corresponding output even when that input pattern has never been seen. Euclidean ARTMAP neural network or better known as EARTMAP neural network is trained according to a certain algorithm that calculates the Euclidean distance and decides to whether include the new pattern in an already existing category (cluster) and update its position in the clustering map, or to consider it as a new category if it is far enough from all of the existing categories. The new location of a cluster center is found by averaging the location of all of the patterns that belong to the cluster itself. This would help in suppressing the white noise level that accompanies those patterns during training. The above mentioned algorithm is tested in a control experiment and worked as a human like system to track a moving target in the plane. The importance of EARTMAP neural network is its ability to imitate certain systems to give a performance that is close to the original performance with a minimum number of categories.

Fuzzy systems are modular by definition but lack the trainability of neural networks. We will introduce a modular, neural-fuzzy system called threshold fuzzy systems (TFS). A TFS has two unique features that distinguish it from traditional fuzzy logic and neural network systems; (1) the rulebase of a TFS contains only single antecedent, single consequence rules (called a behaviorist fuzzy rulebase (BFR)) and (2) the fuzzy inference mechanism is modified to incorporate a highly structured adaptive node network (called a rule dominance network - RDN). Each rule in the BFR is a direct mapping of an input sensor to a system output. Connection nodes in the DN occur when rules in the BFR are conflicting. The nodes of the DN contain functions that are used to suppress the output of other conflicting rules in the BFR. Several different approaches to tuning the unknown parameters of the dominance function can be used; including supervisory training methods, self-organizing and evolutionary-based exploration. For the supervisory training approach, a unique gradient-matrix error back-propagation algorithm (GMEBA) has been developed and will be discussed.

We demonstrate that an omni-directional vehicle (ODV) with an intelligent motion planning (IMP) scheme and a combination of Ackerman & parallel steering capabilities can readily maneuver in constricted areas. A virtual simulation environment (VSE) has been developed to model the ODV motion, a suite of navigation sensors, surrounding objects and the IMP scheme. The IMP unifies the navigation sensors to enhance manual, semi-autonomous and autonomous operations. The VSE is used to demonstrate that the ODV with IMP can navigate tight areas around obstacles, saving energy and time. Metrics for performance of Ackerman versus parallel steering capabilities are also evaluated.

A description is given of a cognitive preview control strategy for autonomous-vehicle steering and cruise guidance by combining optimal preview control theory with rule-based perceptive cruise command generation. The authors also propose a self-training cognition procedure for determining a suitable perceptive schedule for cognitive cruise and steering control. The control yields humanlike driving action in path navigation. It is an intelligent control that decides the cruising speed, plans its control action, and learns the limitation of its steering control. The strategy is being simulated and tested on an autonomous robotic vehicle testbed which is designed for intelligent control experimentation. >

Introduces a trainable control methodology called threshold fuzzy systems (TFS). The need for TFS methods comes from controlling plants that exhibit a varying priority hierarchy. A trademark of these variable priority systems is that if a lower priority system task is completed before a higher priority one the system could become unstable. Designing a control structure that will take into account the varying priorities of a system is difficult. TFS control is shown to be a good approach to these types of systems. Key to the proposed TFS structure is the inclusion of a behaviorist fuzzy rulebase (BFR) and a rule dominance mechanism (RDM) into the classical fuzzy control architecture. The purpose and uses of the BFR and RDM are explained. The TFS control scheme is presented and symbolic and numerical examples are presented.

The jumbo container cranes at the European Container Terminal in Rotterdam are subject of a long term study to automate the international seaport and make container handling more efficient. In the paper appropriate details of the 3D dynamical behavior of the container is modeled. Simulation and visualization are carried out to study the crane behavior under disturbing conditions like unbalanced center of gravity and varying side winds. The software is used to evaluate time-optimal trajectories and control schemes for the dynamics of the electrical drives, and of the swing and skew of the container.

This paper describes a laboratory project for teaching the fundamental concepts of electrohydraulic servomechanism to undergraduate/graduate control engineering students. Although electrohydraulic servomechanisms have dominated the industrial control scene for quite some time, typical control courses do not have these as an integral part of the overall laboratory experience. Thus, the main objective of the proposed experiments is to alleviate this deficiency to a certain extent. The paper includes detailed descriptions of the experimental setup and the laboratory projects.

Computer visualization is an extremely useful tool for conveying engineering concepts to technical as well as non-technical audiences. Real-Time Simulation and Animation (RTSA) is a software that has proven to be very effective in presenting concepts of dynamic control systems in instructional and research laboratories. The unique feature of this recently developed software is that it graphically animates the physical motion of the control system being evaluated, instead of merely performing numerical simulation and displaying plots of the simulated results. With suitable hardware and programming, these tasks can all be accomplished in real-time, giving the visual impact of happening "live". RTSA has been successfully developed and used at Oakland University to provide an interactive computer-based learning experience for control engineering students. It enhances the control curriculum in the following ways: It provides an effective method for introducing applications and concepts of control engineering in real-world systems, and it challenges students to analyze, design and modify the control systems using the interactive features of RTSA software. Instant gratification in seeing control theory successfully applied and better understanding of control concepts have led to the wide acceptance of the RTSA by students. This paper describes the details of how one such RTSA software, entitled "Auto-Tracking Control of a Heat-Seeking Missile," becomes a hit as a laboratory teachware among the students.

In recent years, there has been a growing interest in controlling both active and semi-active automotive suspension systems with a goal of improving ride comfort and vehicle handling. Many such resulting approaches have used linearized models Of the syspension's dynamics, allowing th use of linear (optimal) control theory. In actuality through, these systems and their optimal control are quite nonlinear. In this paper we propose a novel, yet highly practical alternative to such linearized design methods. This alternate optimal design method consists of a modified A* optimal-path, farward-search algorithm which is highly efficient, together with neural networks. The A* search, using a reasonably accurate system model and a given cost function, establishes te nonlinear optimal parametric control Of the suspension. The neural network, as will be shown, learns this nonlinear optimal control function, and in many ways outperforms the search from which it was taught.

A nonlinear Self-Tuning Control (STC) scheme for the nonlinear vehicle active suspension system is investigated and analyzed as an alternative suspension control methodology. The result of the nonlinear STC approach is compared with the linear STC and the model reference adaptive control (MRAC) schemes. The objective of these adaptive control active suspension systems (ACASS) is to tune the controller of the active suspension system automatically such that vehicle disturbance and vibration are reduced to a prespecified level determined by an "ideal" conceptual suspension system. Thus, an ACASS can accommodate for variances in sprung load and suspension component characteristics. The design and performance analysis of a nonlinear STC suspension system is based on a quarter-car model. Our results show significant improvement in the road handling performance of a high-speed vehicle with the STC active suspension system.