Perimeter Patrol Research Articles

Coordination of Robot Teams Over Long Distances: From Georgia Tech to Tokyo Tech and Back-An 11,000-km Multirobot Experiment

One reason why multirobot systems are preferred, in some contexts, over their single-robot counterparts is that they admit a larger spatial footprint. With multiple robots, it is possible to span a large area, which is desirable in applications such as search and rescue, environmental monitoring, and perimeter patrol [1]. However, as interrobot distances grow, so do the communication delays experienced among the robots, making precise, coordinated control hard to achieve [2]-[5].

Multi-robot inverse reinforcement learning under occlusion with estimation of state transitions

Inverse reinforcement learning (IRL), analogously to RL, refers to both the problem and associated methods by which an agent passively observing another agent's actions over time, seeks to learn the latter's reward function. The learning agent is typically called the learner while the observed agent is often an expert in popular applications such as in learning from demonstrations. Some of the assumptions that underlie current IRL methods are impractical for many robotic applications. Specifically, they assume that the learner has full observability of the expert as it performs its task; that the learner has full knowledge of the expert's dynamics; and that there is always only one expert agent in the environment. For example, these assumptions are particularly restrictive in our application scenario where a subject robot is tasked with penetrating a perimeter patrol by two other robots after observing them from a vantage point. In our instance of this problem, the learner can observe at most 10% of the patrol.We relax these assumptions and systematically generalize a known IRL method, Maximum Entropy IRL, to enable the subject to learn the preferences of the patrolling robots, subsequently their behaviors, and predict their future positions well enough to plan a route to its goal state without being spotted. Challenged by occlusion, multiple interacting robots, and partially known dynamics we demonstrate empirically that the generalization improves significantly on several baselines in its ability to inversely learn in this application setting. Of note, it leads to significant improvement in the learner's overall success rate of penetrating the patrols. Our methods represent significant steps towards making IRL pragmatic and applicable to real-world contexts.

Lower Bounding Linear Program for the Perimeter Patrol Optimization Problem

In this article, a stochastic optimal control problem involving an unmanned aerial vehicle flying patrols around a perimeter is considered. To determine the optimal control policy, one has to solve a Markov decision problem, whose large size renders exact dynamic programming methods intractable. Therefore, a state aggregation based approximate linear programming method is used instead, to construct provably good suboptimal patrol policies. The state space is partitioned and the optimal cost-to-go or value function is restricted to be a constant over each partition. The resulting restricted system of linear inequalities embeds a family of Markov chains of lower dimension, one of which can be used to construct a lower bound on the optimal value function. In general, the construction of a lower bound requires the solution to a combinatorial problem. But the perimeter patrol problem exhibits a special structure that enables tractable linear programming formulation for the lower bound. This is demonstrated and numerical results that corroborate the efficacy of the proposed methodology are also provided.

Optimization of Perimeter Patrol Operations Using Unmanned Aerial Vehicles

This paper addresseses the following base perimeter patrol problem: a team of unmanned aerial vehicles (UAVs) equipped with cameras and a remotely located operator cooperatively perform the task of perimeter surveillance. There arem alert stations/sites on the perimeter where a nearby breaching of the perimeter by an intruder is flagged byanunattended ground sensor (UGS).Todeterminewhether an incursionflaggedby aUGS is a false alarmora real threat, a patrolling UAV flies to the alert site to investigate the alert. The longer a UAVdwells (loiters) at an alert site, the more information it gathers; however, this also increases the delay in responding to other alerts. The decision problem for aUAV is to determine the optimal dwell time so as tomaximize the expected payoff. In this paper, patrols consisting of one and two UAVs are considered. A stochastic dynamic programming approach is employed to obtain optimal policies for the patrolling UAVs. Theoretical performance bounds from queueing systems literature have been used to benchmark the optimal controller. Also, simulation results for the optimal patrols showing the expected information gained and response time for different alert arrival rates are presented.

Adaptive Immune System TH1/TH2 Differentiation Mechanism Inspired Perimeter Patrol Control Strategy

This brief investigates a perimeter patrol control problem in which multiple agents travel back and forth defending a border. A decentralized control inspired from the adaptive immune system TH1/TH2 differentiation mechanism has been proposed. The resulting patrol control exhibits robustness under adversarial conditions. This is primarily because the motion period can be arbitrary, which makes it difficult for intruders to find weak spots for attack. Moreover, this immuno-inspired control can avoid collision and automatically recover from agent failure.

Approximate dynamic programming with state aggregation applied to UAV perimeter patrol

AbstractOne encounters the curse of dimensionality in the application of dynamic programming to determine optimal policies for large‐scale controlled Markov chains. In this paper, we provide a reward‐based aggregation method to construct suboptimal policies for a perimeter surveillance control problem which gives rise to a large scale Markov chain. The novelty of this approach lies in circumventing the need for value iteration over the entire state space. Instead, the state space is partitioned and the value function is approximated by a constant over each partition. We associate a meta‐state with each partition, where the transition probabilities between these meta‐states are known. The state aggregation approach results in a significant reduction in the computational burden and lends itself to value iteration over the aggregated state‐space. We provide bounds to assess the quality of the approximation and give numerical results that support the proposed methodology. Published 2011. This article is a US Government work and is in the public domain in the USA.

Human vs. Algorithmic Path Planning for Search and Rescue by Robot Teams

Substantial automation will be needed to allow operators to control the large teams of robots envisioned for search and rescue, perimeter patrol, and a wide variety of military tasks. Both analysis and research point to navigation and path planning as prime candidates for automation. When operators are isolated from robot navigation, however, there may be loss of situation awareness (SA) and difficulties in monitoring robots for failures or abnormal behavior. Operator's navigational strategies are quite complex and extremely changeable at foraging tasks in unknown environment reflecting background knowledge and expectations about human and natural environments. These considerations are missing from automated path planning algorithms leading to differences in search patterns and exploration biases between human and automatically generated paths. Effectively integrating automated path planning into multirobot systems would require demonstrating that: 1-automated path planning performs as well as humans on measures such as area coverage and 2- use of automated path planning does not degrade performance of related human tasks such as finding and marking victims. In this paper we seek to compare the divergence between human manual control and autonomous path planning at an urban search and rescue (USAR) task using fractal analysis to characterize the paths generated by the two methods. Area coverage and human contributions to mixed-initiative planning are compared with fully automated path planning. Finally, the impact of automated planning on related victim identification and marking tasks is compared for automated paths and paths generated by previous participants.

Security PIDS with physical sensors, real-time pattern recognition, and continuous patrol

This perimeter detection system (PIDS) is a system applicable to surveillance and perimeter control of areas of responsibility where risks are assessed, like swimming-pools, school precincts, museums, embassies, warehouses, and the like. It consists of a string of an arbitrary number of spaced but communicated micro-controllers, each one driving infra-red (IR) and/or ultrasound emitters and digital output transducers. The system uses radar effect and it carries continuous perimeter patrol by performing sequential surveillance of detector outputs as well as distributed pattern recognition analysis. All micro-controllers support a common code which includes turning the emitters sequentially on and off, pattern recognition routines, and serial transmission of alarm byte. Butterfly alarms are avoided via software and adequate geometrical configuration of the emitter beam net. One master micro-controller closes the string synchronizes and conducts the others. When the master micro-controller receives the intruder signal, it triggers the physical alarm, its next order being to restart patrolling.