Majura F. Selekwa

Robot navigation in very cluttered environments by preference-based fuzzy behaviors

One of the key challenges in application of Autonomous Ground Vehicles (AGVs) is navigation in environments that are densely cluttered with obstacles. The control task becomes more complex when the configuration of obstacles is not known a priori. The most popular control methods for such systems are based on reactive local navigation schemes that tightly couple the robot actions to the sensor information. Because of the environmental uncertainties, fuzzy behavior systems have been proposed. The most difficult problem in applying fuzzy-reactive-behavior-based navigation control systems is that of arbitrating or fusing the reactions of the individual behaviors, which is addressed here by the use of preference logic. This paper presents the design of a preference-based fuzzy behavior system for navigation control of robotic vehicles using the multivalued logic framework. As shown in simulation and experimental results, the proposed method allows the robot to smoothly and effectively navigate through cluttered environments such as dense forests. Experimental comparisons with the vector field histogram method (VFH) show that the proposed method usually produces smoother albeit longer paths to the goal.

The virtual wall approach to limit cycle avoidance for unmanned ground vehicles

Robot Navigation in unknown and very cluttered environments constitutes one of the key challenges in unmanned ground vehicle (UGV) applications. Navigational limit cycles can occur when navigating (UGVs) using behavior-based or other reactive algorithms. Limit cycles occur when the robot is navigating towards the goal but enters an enclosure that has its opening in a direction opposite to the goal. The robot then becomes effectively trapped in the enclosure. This paper presents a solution named the Virtual Wall Approach (VWA) to the limit cycle problem for robot navigation in very cluttered environments. This algorithm is composed of three stages: detection, retraction, and avoidance. The detection stage uses spatial memory to identify the limit cycle. Once the limit cycle has been identified, a labeling operator is applied to a local map of the obstacle field to identify the obstacle or group of obstacles that are causing the deadlock enclosure. The retraction stage defines a waypoint for the robot outside the deadlock area. When the robot crosses the boundary of the deadlock enclosure, a virtual wall is placed near the endpoints of the enclosure to designate this area as off-limits. Finally, the robot activates a virtual sensor so that it can proceed to its original goal, avoiding the virtual wall and obstacles found on its way. Simulations, experiments, and analysis of the VWA implemented on top of a preference-based fuzzy behavior system demonstrate the effectiveness of the proposed method.

Online Terrain Classification for Mobile Robots

Today’s autonomous vehicles operate in an increasingly general set of circumstances. In particular, unmanned ground vehicles (UGV’s) must be able to travel on whatever terrain the mission offers, including sand, mud, or even snow. These terrains can affect the performance and controllability of the vehicle. Like a human driver who feels his vehicle’s response to the terrain and takes appropriate steps to compensate, a UGV that can autonomously perceive its terrain can also make necessary changes to its control strategy. This article focuses on the development and application of a terrain detection algorithm based on terrain induced vehicle vibration. The dominant vibration frequencies are extracted and used by a probabilistic neural network to identify the terrain. Experimental results based on iRobot’s ATRV Jr (Fig. 1) demonstrate that the algorithm is able to identify with high accuracy multi-differentiated terrains broadly classified as sand, grass, asphalt, and gravel.Copyright

Path tracking control of four wheel independently steered ground robotic vehicles

Because of their high degree of maneuverability, four wheel steered robotic vehicles have increasingly attracted interest in many applications. These robots are characterized by ability to easily maneuver tight turns. Two types of such vehicles have been identified: vehicles with independently steered wheels, and vehicles with mechanically coupled pairs of steered wheels. While the latter group has been easy to model and control, the former group still poses many control challenges. The common approaches employed in modeling and controlling vehicles with independently steered wheels either assume that the rear wheels will copy the front steering angles or allow wheel slippage to accommodate independent steering angles for all wheels. Both these approaches do not offer the sought maneuverability advantages. This paper revisits the problem of four independent wheel steering and proposes an approach that makes it possible to achieve maximum maneuverability while avoiding wheel slippage. It develops individual wheel constraints that enable the wheels to be independently controlled while satisfying the desired vehicle motion. Numerical simulation results have shown that this approach can indeed simplify the problem of controlling four steered wheel vehicles.

A fuzzy logic approach to LQG design with variance constraints

One of the well-known deficiencies of most modern control methods (H/sub 2/, H/sub /spl infin//, and L/sub 1/ designs) is that they attempt to represent multiple criteria with scalar cost functions. Hence, in practice the (static or dynamic) weights in the scalar cost function must be determined by an iterative process in order to satisfy the multiple objectives. This paper develops a fuzzy algorithm for selecting the weights in a linear quadratic Gaussian (LQG) cost functional such that the constraints on the variances of the system are satisfied. This problem is denoted as a variance constrained LQG problem. Variations of this problem are considered in the existing literature using crisp logic. Numerical experiments show that when both the input and output variances are constrained, the fuzzy algorithm converges faster and tends to be much more robust to new systems or constraints than the crisp algorithms.

Robust fault detection using robust l/sub 1/ estimation and fuzzy logic

The paper considers the application of robust l/sub 1/ estimation in conjunction with fuzzy logic to robust fault detection for an aircraft flight control system. It reviews the design of robust l/sub 1/ estimators based on multiplier theory and the resulting fixed threshold approach to fault detection (FD). It also discusses the principles behind fuzzy logic as applied to robust residual evaluation and FD. Due to modeling errors and unmeasurable disturbances, it is difficult to distinguish between the effects of an actual fault and those caused by uncertainty and disturbance. Hence, it is the aim of a robust FD system to be sensitive to faults while remaining insensitive to uncertainty and disturbances. While fixed thresholds only allow a decision on whether a fault has or has not occurred, it is more valuable to have the residual evaluation lead to a conclusion related to the degree of, or probability of, a fault. Fuzzy logic is a viable means of determining the degree of a fault and allows the introduction of human observations that may not be incorporated in the rigorous threshold theory. Hence, fuzzy logic can provide a more reliable and informative fault detection process. Using an aircraft flight control system, the results of FD using robust l/sub 1/ estimation with a fixed threshold are demonstrated. FD that combines robust l/sub 1/ estimation and fuzzy logic is also demonstrated. It is seen that combining the robust estimator with fuzzy logic proves to be advantageous in increasing the sensitivity to smaller faults while remaining insensitive to uncertainty and disturbances.

Centralized fuzzy behavior control for robot navigation

Simplicity, flexibility, and responsiveness are among the reasons that behavior fuzzy control in robotic systems is popular. One of the major problems faced in the design of behavior based fuzzy control systems for mobile robots is that of resolving conflicts between different behaviors when the behaviors are given full autonomy, i.e., the control is decentralized. Under this framework the behaviors are forced to compete for robot control whereby some of the behaviors must lose the competition and hence be ignored by the robot. By ignoring the behaviors that lost in the competition, the control system becomes less robust and can lead the robot to undesired reactions. This paper presents a new method of designing fuzzy behavior control systems for robotic navigation that honors all behaviors. The control system is centralized and it takes into consideration the interests of all behaviors. Each behavior is designed to fire several proposals on how to react to a given situation. There is a central command unit that takes the proposals from all behaviors in the system and selects a single proposal that best meets the demands of all behaviors in the system. This structure allows the behavioral fusion to be robust in the sense that under all circumstances, each behavior will be honored and the robot reactions will satisfy the requirements of each behavior.

Fuzzy behavior navigation for an unmanned helicopter in unknown environments

Aerial missions that require unmanned aerial vehicles (UAVs) to fly autonomously in unknown and hostile environments are inevitable. These UAVs must be equipped with a fully autonomous navigation system. Many methods that have been proposed for navigation of autonomous systems either lack the necessary intelligence or are not responsive enough to cope with the flying speeds of UAVs. This paper presents a new method for autonomous navigation of UAVs using reactive fuzzy behaviors. It extends a 2D fuzzy behavior navigation system used in unmanned ground vehicles to a 3D navigation system suitable for UAVs. The research is based on a range finding sensor system by judicious use of a 2D range finder. A novel defuzzification method for 3D navigation systems is developed to generate the most desired flying orientation. Simulation results carried out using MATLABs Virtual Reality Toolbox show that the proposed system works very well to avoid static obstacles.

/spl Hscr//sub 2/ optimal reduced order control design using a fuzzy logic methodology with bounds on system variances

There are practical reasons for controlling systems in such a way that variances of selected inputs and outputs are kept within specified limits. One way of designing control laws that achieve these objectives is by using a linear quadratic Gaussian (LQG) (or optimal /spl Hscr//sub 2/) control design method with an appropriate choice of the input and output weighting matrices. Since the LQG controller is of the same order as the plant being controlled, its practical implementation tends to be very difficult for higher order plants unless the controller order is reduced. This letter considers the design of /spl Hscr//sub 2/ optimal reduced order controllers to meet a set of variance constraints. This problem also involves the proper choice of the weighting matrices in the cost function. The fuzzy algorithm previously developed for the full-order variance constrained problem is shown to be applicable to the reduced order variance constrained problem. Three reduced order design schemes are developed and compared. Two schemes involve direct reduced order design and one scheme involves reduced order design using modified balanced truncation. The three schemes are compared using numerical experiments. The results clearly demonstrate the feasibility of reduced order /spl Hscr//sub 2/ optimal design that satisfy variance constraints on the system.

SETTING UP A PROBABILISTIC NEURAL NETWORK FOR CLASSIFICATION OF HIGHWAY VEHICLES

Many neural network methods used for efficient classification of populations work only when the population is globally separable. In situ classification of highway vehicles is one of the problems with globally nonseparable populations. This paper presents a systematic procedure for setting up a probabilistic neural network that can classify the globally nonseparable population of highway vehicles. The method is based on a simple concept that any set of classifiable data can be broken down to subclasses of locally separable data. Hence, if these locally separable data can be identified, then the classification problem can be carried out in two hierarchical steps; step one classifies the data according to the local subclasses, and step two classifies the local subclasses into the global classes. The proposed approach was tested on the problem of classifying highway vehicles according to the US Federal Highway Administration standard, which is normally handled by decision tree methods that use vehicle axle information and a set of IF-THEN rules. By using a sample of 3326 vehicles, the proposed method showed improved classification results with an overall misclassification rate of only 2.9% compared to 9.7% of the decision tree methods. A similar setup can be used with different neural networks such as recurrent neural networks, but they were not tested in this study especially since the focus was for in situ applications where a high learning rate is desired.