A Framework for Generalized Risk-Aware Planning and Decision-Making under Dynamic Adversarial Conditions

Abstract: Path planning in dynamic and adversarial environments has been one of the most crucial challenge in the world of artificial intelligence and robotics. Traditional algorithms such as A* and ARA* primarily aimed at achieving the optimized path often neglecting the adversarial threats and safety concerns. Addressing this limitation, in this paper we have proposed the ARA++, a novel framework that extends the ARA* algorithm by integrating probabilistic enemy risk factor modeling, adaptive thresholding and multi-objective cost functions. The ARA++ algorithm unlike conventional path finding algorithms evaluates risk factor and survival probability for nodes at each time interval providing dynamic updates of the risk factors on the basis of adversary movement. A safety threshold value must be defined for the problem such that if any path’s survival probability value falls below the safety threshold, the path is pruned so that only safe trajectories are considered. A case study of 3*3 maze problem is considered that clearly demonstrates the working of algorithm and finding the shorter as well as safer path maximizing the probability of survival. Experimental analysis confirms that ARA++ outperforms the conventional algorithms A* and ARA* in adversarial settings by producing safer, cheap, adaptive and computationally efficient paths. Extending the limits from a simple maze problem, the algorithm is generalized to be used across different fields that can be automated vehicles, robotics and cybersecurity, thus, making it an entire framework.