This paper describes initial results regarding navigation and control algorithms developed for Gamera, an autonomous underwater vehicle. In order to perform high precision controlled motions, the robot employs a novel propulsion system known as Nektons, and carries a comprehensive sensor suite to track orientation and position. A model of the robot has been developed that relates the 12 robot controlled propulsion outputs-frequency, amplitude, and center point of oscillation for each of the four motors-to the forces generated by the entire system. We explore the rationalization of control algorithms for this system. Using these algorithms, Gamera demonstrates a high level of control and successfully performs pre-specified maneuvers autonomously.

Through our work we have identified hardware and software components that enable us to mimic biological entities and colonies within the context of multi-agent systems. Specifically, we use control networks with fieldbus or wireless data links to connect distributed hardware and software modules. Although there are many implementations of multi-agent systems, most literature seems to focus on the algorithmic aspects. We describe how (LonWorks) control networks, coupled with minimal custom electromechanical devices, can be used to construct multi-agent systems. Important issues include real-time system constraints, fault-tolerance, synchronization, and, of course, solving the robotic tasks at hand. We describe three proof-of-concept projects currently underway: a biped walking robot; a hexapod colony; and a complex autonomous vehicle. It is believed that attributes observed in these projects can be used to build application-specific systems in other areas.

We fuse a self-organizing hyperellipsoid clustering (HEC) Kohonen neural network with the Julier-Uhlmann-Kalman filter (JUKF) to perform map building and low-level position estimation. The HEC Kohonen uses the Mahalanobis distance to learn elongated shapes (typical of sonar data) and obtain a stochastic measurement of data-node association. The number of nodes is regulated by measuring how well a node model matches its associated data. The HEC Kohonen can handle high-dimensional problems and can be generalized to other pattern recognition problems. The JUKF compliments the HEC Kohonen in that it performs low-level (nonlinear) tracking more efficiently and more accurately than the extended Kalman filter. By estimating and propagating error covariances through system transformations, the JUKF eliminates the need to derive Jacobian matrices. The inclusion of stochastic information inherent to the HEC map renders the JUKF an excellent tool for our HEC-based map building, position estimation, motion planning and low-level tracking.

Our research centers around finding hardware and software components that enable us to mimic biological entities and colonies within the context of multi-agent systems. Specifically, we use control networks with fieldbus or wireless data links to connect distributed hardware and software modules. Although there are many implementations of multi-agent systems, most literature seems to focus on the algorithmic aspects. In this paper, we describe how LonWorks control networks, coupled with minimal custom electromechanical devices, can be used to construct multi-agent systems. Important issues include real-time system constraints, fault-tolerance, synchronization, and, of course, solving the robotic tasks at hand. We describe three proof-of-concept projects currently underway: a biped walking robot; an hexapod colony; and a complex autonomous vehicle. It is believed that attributes observed in these projects can be used to build application-specific systems for automation, space exploration, hazardous materials operations, etc.

This paper documents how two mobile robots can share knowledge about their environment. The two mobile robots we use have different sensor configurations, drive systems and other physical attributes including weight and size. "Knowledge" is generated by the region-feature neural network (RFNN), and can be transferred on two general levels: (1) a complete transfer of a matured neural network; and (2) a transfer of matured features. This transferred knowledge can also be "tuned" with and without locking the feature level synaptic weights. We examine the impact and feasibility of sharing (on both levels, with and without locking features) neural network modules trained on actual sonar data in the global self-localization (GSL) problem. Significant reductions in training time are realized and presented. We also describe the neural network architecture and our general approach to solving the GSL problem in a time-, translation- and rotation-invariant way.

In this paper we show how a self-organizing Kohonen neural network can use hyperellipsoid clustering (HEC) to build maps from actual sonar data. Since the HEC algorithm uses the Mahalanobis distance, the elongated shapes (typical of sonar data) can be learned. The Mahalanobis distance metric also gives a stochastic measurement of a data point's association with a node. Hence, the HEC Kohonen can be used to build topographical maps and to recognize its own topographical cites for self-localization. The number of nodes can also be regulated in a self-organizing manner by using the Kolmogorov-Smirnov (KS) test for cluster compactness. The KS test determines whether a node should be divided (mitosis) or pruned completely. By incorporating principal component analysis, the HEC Kohonen can handle problems with several dimensions (3D, time-series, etc.). The HEC Kohonen is also multifunctional in that it accommodates compact geometric motion planning and self-referencing algorithms. It can also be used to solve a host of other pattern recognition problems.

Approaches in global motion planning (GMP) and geometric beacon collection (for self-localization) using traversability vectors have been developed and implemented in both computer simulation and actual experiments on mobile robots. Both approaches are based on the same simple, modular, and multifunctional traversability vector (t-vector). Through implementation it has been found that t-vectors reduce the computational requirements to detect path obstructions, Euclidean optimal via-points, and geometric beacons, as well as to identify which features are visible to sensors. Environments can be static or dynamic and polygons are permitted to overlap (i.e., intersect or be nested). While the t-vector model does require that polygons be convex, it is a much simpler matter to decompose concave polygons into convex polygon sets than it is to require that polygons not overlap, which is required for many other GMP models. T-vectors also reduce the data size and complexity of standard V-graphs and variations thereof. This paper presents the t-vector model so that the reader can apply it to mobile robot GMP and self-localization.

Autonomous vehicles require that all processes be efficient in time, complexity and data storage. In fact, an ideal system employs multifunctional models where ever possible. This paper presents the region- and feature-based neural network (RFNN) as a viable pattern analysis process engine for solving a variety of problems with a single math model. The RFNN employs receptive fields and weight sharing which compensate for noise, minor phase shifts and occlusions. The RFNN also utilizes greedy adaptive learning rates and mature feature preservation to expedite the overall training process. A novel ad hoc approach called "shocking" is used to solve the instability problem inherent to greedy adaptive learning rates. The basic RFNN "feature" is grounded in computer vision morphology in that the neural network autonomously learns subpatterns unique to various problems. This paper comprehensively describes the flexible RFNN architecture and training process and presents two problems that can be solved by the RFNN: sensor pattern-recognition and traffic sign recognition.

An approach to global motion planning for autonomous mobile robots has been developed on the basis of traversability vectors (t-vectors). Through the overall course of this research it was found that t-vectors provide a utility, efficiency and mathematical stability for collision detection and visibility that cannot be matched by commonly used algebraic approaches in static and dynamic environments. This paper will show that t-vectors also impact global motion planning by identifying redundancies in visibility graphs (V-graphs) and expediting their construction. The result of eliminating redundant path segments is a streamlined version of the V-graph called the essential visibility graph (EVG). This paper will also show that the EVG offers a significant reduction in data storage requirements and complexity.

An approach to global self-localization for autonomous mobile robots has been developed using self-organizing Kohonen neural networks. This approach categorizes discrete regions of space using mapped sonar data corrupted by noise of varied sources and ranges. Our approach is similar to optical character recognition (OCR) in that the mapped sonar data can, over time, assume the form of a character unique to that room. Hence, it is believed that an autonomous vehicle can be capable of determining which room it is in based on mapped sensory data ascertained by wandering through and exploring that room. With some pre-processing and a robust explore routine, the solution becomes time-, translation- and rotation-invariant.

A motion planning and self-referencing approach has been developed, simulated and applied to an actual robot. Although there are several novelties to these approaches, the fact that both are based on traversability vectors (t-vectors) is one aspect of this research that is unique. Through their application it has been found that t-vectors enhance the detection of path obstructions and geometric beacons and expedite the identification of features that are visible (or hearable) to sensors in both static and dynamic environments. T-vectors also reduce the data size and complexity of standard V-graphs and variations thereof. This paper provides the t-vector models step-by-step so that the reader will be able to apply them to mobile robot motion planning and self-referencing.>

When navigating through an environment, a mobile robot updates its position and orientation by searching for known objects called geometric beacons. This operation of comparing predicted sonar ranges with observed ones is referred to as self-referencing. There are two fundamental steps necessary for solving the self-referencing problem: how to represent the objects in the environment and how to search for landmarks based on this representation. This paper presents a method of mobile robot self-referencing where every object in the environment can be used as a landmark beacon. First, it shows how to interpret and predict sonar range readings through regional sampling and sensor window construction. Second, it justifies geometrically representing obstacles on the basis of surface information quality, compatibility with global motion planning, and the computer storage necessary to satisfy both criteria. Finally, it justifies the use of traversability vectors and configuration-space-time based on how they can be simultaneously used to plan motion globally and facilitate the search for geometric beacons.