Catálogo de publicaciones - libros

We introduce the team dispatching (TD) problem arising in cooperative control of multiagent systems, such as spacecraft constellations and UAV fleets. The problem is formulated as an optimal control problem similar in structure to queuing problems modeled by restless bandits. A near-optimality result is derived for greedy dispatching under oversubscription conditions, and used to formulate an approximate deterministic model of greedy scheduling dynamics. Necessary conditions for optimal team configuration switching are then derived for restricted TD problems using this deterministic model. Explicit construction is provided for a special case, showing that the most-oversubscribed-first (MOF) switching sequence is optimal when team configurations have low overlap in their processing capabilities. Simulation results for TD problems in multi-spacecraft interferometric imaging are summarized.

We describe a pursuer-evader game played on a grid in which the pursuers can move faster than the evaders, but the pursuers cannot determine an evader’s location except when a pursuer occupies the same grid cell as that evader. The pursuers’ object is to locate all evaders, while the evader’s object is to prevent collocation with any pursuer indefinitely. The game is loosely based on autonomous unmanned aerial vehicles (UAVs) with a limited field-of-view attempting to locate enemy vehicles on the ground, where the idea is to control a fleet of UAVs to meet the search objective. The requirement that the pursuers move without knowing the evaders’ locations necessitates a model of the game that does not explicitly model the evaders. This has the positive benefit that the model is independent of the number of evaders (indeed, the number of evaders need not be known); however, this has the negative side-effect that the time and memory requirements to determine a pursuer-winning strategy is exponential in the size of the grid. We report significant improvements in the available heuristics to abstract the model further and reduce the time and memory needed.

The authors have performed formation flights of UAVs using two of UC Berkeley’s BEAR unmanned helicopters together in realtime with a simulated leader and six simulated helicopters. The goal of this experiment was to verify the mesh stability theory. The experimental results differ from the ideal theoretical results. Upon closer examination, this discrepancy is due to the effects of having a heterogeneous formation while the theory used is meant for homogeneous formations. While much work remains to be done, we can still show that using leader information in the control law is better than not using leader information. However, a new question arises regarding how one should define mesh stability for a heterogeneous mesh.

A navigation algorithm using Time Difference of Arrival (TDOA) measurements obtained from signals of opportunity (SOPs) is developed. SOP-derived TDOA measurements are signals that are transmitted for purposes other than navigation (such as communication, telecasts, etc.) and are motived as appealling alternatives to GPS. In the scenario considered herein, the received SOPs are generated by asynchronous emitters at known locations. The measured TDOA are with reference to a reference receiver also at a known location. Although it would appear that the equations governing TDOA measurements, and consequently TDOA GDOP, are quite different from their GPS counterparts, it is shown that the TDOA measurement equations can be transformed into GPS pseudorange equations. This interpretation not only provides a direct evaluation of the TDOA GDOP and affords a characterization of the optimal measurement geometry, but, in addition, lets us fall back on well known GPS algorithms.

In this chapter, we present a comparative study on two algorithms to localize ground targets using Unmanned Aerial Vehicles (UAVs): an angle-of-arrival (AOA) emitter-location algorithm using triangulation techniques and an angle-rate algorithm. In particular, we focus on the performance of the two algorithms locating targets when a sensor platform is under a large Geographic Dilution of Precision (GDOP) condition. The large GDOP condition occurs when a target is seen by a sensor platform within a small included angle; the total included angle between Line-Of-Bearings (LOBs) is less than five degrees. The comparative study is a part of the United States Air Force Academy’s Unmanned Aerial Vehicles (UAVs) research project to develop a group of cooperative UAVs to search, detect, and localize moving ground targets. The GDOP conditions limit the accuracy of the AOA triangulationemitter-location algorithm’s accuracy due to the resulting highly elliptical probable error. In such cases, angle-rate algorithms should be used for better localization accuracy. Usually, a large GDOP condition is encountered during two important operational applications: (1) tasks that use slow-moving-sensor platforms, such as a small UAV, and (2) tasks involving short-up-time emitters that typically are not transmitting signals long enough for any sensor platform to open more than a small total included angle. We investigate the performance of the two algorithms as we vary the included angle for the sensor platform. The performances of angle-rate and triangulation algorithms are compared via MATLAB simulation to determine the preferred regions of operation.

In this paper, a new leaderless cooperative formation control strategy is proposed for a group of autonomous mobile robots. Through the local state and input transformations, the formation control problem can be recast as the cooperative control design problem for a class of general canonical systems with arbitrary but finite relative degree. A set of less-restrictive sufficient conditions on group communication topology to ensure the success of cooperative control design has been established. The system stability is rigorously proved by studying the convergence of products of row stochastic matrices. The proposed design does not require either that collaborative robots have a fixed communication/control structure (such as leader/follower or nearest neighbor) or that their sensor/communication graph be strongly connected. Detailed simulation results are provided to illustrate the effectiveness of the proposed method.

We consider the problem of controlling a system of many Unmanned Air Vehicles (UAVs) whose mission is to patrol and protect a set of important assets on the ground. We present two widely differing methods, employing emergent intelligent swarms and closed-form optimization. The optimization approach assumes complete communication of all newly sensed information among all of the UAVs as it becomes available. The optimization problem is a network flow model that is readily solvable to obtain optimum task allocations to configure patrols for the UAVs in the swarm. Reapplication of the optimization algorithm upon demand yields the benefit of cooperative feedback control. The swarm procedure establishes patrol patterns by utilizing decentralized, reactive, behaviors. Global communication is unnecessary, and control is established only through passive sensors and minimal short-range radio communication. Both models have been implemented and successfully demonstrated in an agent-based, simulated environment. The strengths, weaknesses, and relative performance the two approaches are compared and discussed.

In many cooperative control methods, the geometric state of the system is abstracted to the underlying graph or . In this paper we present a grammatical approach to modeling and controlling the network topology of cooperative systems based on graph rewriting. By restricting rewrites to small subgraphs, provide a useful method for programming the concurrent behavior of large decentralized systems of robots. We illustrate the modeling process through an ongoing example and demonstrate mathematical tools for reasoning about the system’s behavior. Finally, we briefly describe methods to design continuous controllers that augment the grammar so that geometric requirements may also be satisfied.

This chapter describes a distributed system for collaboration and control of a group of unmanned aerial vehicles (UAVs). The system allows a group of vehicles to work together to accomplish a mission via an allocation mechanism that works with a limited communication range and is tolerant to agent failure. This system could be used in a number of applications including mapping, surveillance, search and rescue operations.

The user provides a mission plan containing a set of tasks and an obstacle map of the operating environment. An estimated mission state, described in a high level language, is maintained on each agent and shared between agents whenever possible. This language represents each task as a set of subtasks. Each subtask maintains a state with information on the subtask status, an agent ID, a timestamp, and the cost to complete the subtask. The estimated mission states are based on each agent’s current knowledge of the mission and are updated whenever new information becomes available. In this chapter, each subtask is associated with a point in space, although the system methodology can be expanded to more general subtask types.

The agents employ a three-layer hierarchical decision and control process. The upper layer contains transition logic and a communication process. The transition logic manages transitions between tasks and between subtasks, which determine the behavior of the agent at any given time. The communication process manages the exchange of mission state information between agents. Among other capabilities, the subtask transition rules provide time-based fault management; if an agent is disabled or stops communicating, others will assume its subtask after a mission-dependent timeout period. The middle layer contains a trajectory planner that uses a modified potential field method to generate a safe trajectory for a UAV based on the obstacle map and the current subtask objective. The lower layer contains a trajectory-tracking controller that produces heading and airspeed commands for the UAV. Properties of the system are analyzed and the methodology is illustrated through an example mission simulation.