Bridging the Gap between Observation and Decision Making: Goal Recognition and Flexible Resource Allocation in Dynamic Network Interdiction

Abstract

Goal recognition, which is the task of inferring an agent's goals given some or all of the agent's observed actions, is one of the important approaches in bridging the gap between the observation and decision making within an observe-orient-decide-act cycle. Unfortunately, few research focuses on how to improve the utilization of knowledge produced by a goal recognition system. In this work, we propose a Markov Decision Process-based goal recognition approach tailored to a dynamic shortest-path local network interdiction (DSPLNI) problem. We first introduce a novel DSPLNI model and its solvable dual form so as to incorporate real-time knowledge acquired from goal recognition system. Then a Markov Decision Process-based goal recognition model along with its dynamic Bayesian network representation and the applied goal inference method is proposed to identify the evader's real goal within the DSPLNI context. Based on that, we further propose an efficient scalable technique in maintaining action utility map used in fast goal inference, and develop a flexible resource assignment mechanism in DSPLNI using knowledge from goal recognition system. Experimental results show the effectiveness and accuracy of our methods both in goal recognition and dynamic network interdiction.

Goal Identification Control Using an Information Entropy-Based Goal Uncertainty Metric

“…For simplicity, we select a reduced Chicago Sketch Road Network [20] expanded from the vertex No.368 to its neighbors and neighbors' neighbors for 5 times, consisting of 51 vertexes and 113 edges. The computations of the RGUR and RGUI are formulated into a BLMIP, and SPLUNI into a BLMIP and solved using the MIP solvers of CPLEX 11.5 and YALMIP toolbox of MATLAB [57].…”

“…The goal recognition problem has been formulated and addressed in many ways, as a graph covering problem upon a plan graph [10], a parsing problem over grammar [11][12][13][14], a deductive and probabilistic inference task over a static or Dynamic Bayesian Network [15][16][17][18][19][20] and an inverse planning problem over planning models [21][22][23][24][25].…”

“…where the importance weights W i should be updated recursively. Within this formulation, following research continues to improve the model's expressiveness and computational attractiveness, as in [18,20,[30][31][32][33][34][35][36]. Notably, research in [33,36] proposes a CSSMs model which allow simple system model definition in the form of planning problem and can reason in large state spaces by using approximate inference.…”

“…The authors in [20] shows a simple yet effective probabilistic goal recognizer based on Markov Decision Process (MDP). The probabilistic goal recognition model is defined by a tuple D = S, s 0 , A, f , G, e, O , where S is a finite and discrete state space; s 0 is the start state of the agent; A is the set of actions; f : S × A → S is a deterministic state transition function; G is a set of goal states; e is the goal termination variable and O is a non-empty observation set.…”

“…where N is the number of possible goals, TP i , TI i and TT i are the true positives, total of true labels, and the total of inferred labels for class i respectively. Equations (20) to (22) show that F-measure is an integration of precision and recall, where precision is used to scale the reliability of the recognized results and recall to scale the efficiency of the algorithm applied in the test dataset. The value of F-measure will be between 0 and 1, and a higher metric value means a better performance.…”