This paper presents S-RGM ^* , a sampling-based approach utilizing a geometric model specifically adapted for path planning in unmanned aerial manipulators designed for pick-and-place applications. As the demand for autonomous aerial systems grows in sectors such as inspection, delivery, and maintenance, efficient trajectory planning for manipulators mounted on unmanned aerial vehicles (UAVs) becomes crucial. The proposed approach leverages a geometric model to enhance the accuracy and reliability of path planning, addressing key challenges related to obstacle avoidance, accessibility, and precise object manipulation. Through simulations, S-RGM ^* demonstrates superior performance compared to traditional methods in navigating complex environments and achieving precise manipulator positioning for pick-and-place tasks. This work represents a significant advancement in enhancing the autonomy and operational scope of unmanned aerial manipulators, paving the way for more versatile and efficient aerial robotic systems.
Power grids have become valuable targets for both physical and cyber-attacks. A huge corpus of research and engineering best practices exist with regard to threat modeling. As robustness alone is considered inadequate to counter today’s attack vectors, resilience has become the focus of research. In this vein, learning agents, e. g., based on deep reinforcement learning, have shown a lot of promise both in deriving attack vectors and providing resilience against those. Pure model-free approaches would start training from scratch and need to re-discover already known attacks. Offline reinforcement learning provides a suitable alternative to this. However, the generation of an offline training dataset that reflects the threat model at hand while still providing an extensive dataset to train on is a research gap. Moreover, power grid simulations usually feature time series, which introduce stationarity issues that could invalidate an agent’s policy. In this paper, we apply state machines as tools to generate data based on a well-known attack. We further show how to verify the impact of the environment’s non-stationarity.
We propose a hybrid approach combining population- and trajectory-based metaheuristics to solve continuous optimization problems efficiently. We utilize a Self-Adaptive Binary Space Partitioning (SA-BSP) tree to divide the search space of continuous problems, guiding the hybrid framework toward the most promising subregions. To address the issue of premature convergence, we implement a “Finder-Tracker agents” mechanism. The hybrid framework is structured in three main phases. In the first phase, the SA-BSP tree serves as a memory unit within the population-based algorithm, capturing critical information about the explored areas, building the fitness landscape, and partitioning the search space. In the second phase, an intelligent controller is introduced to balance exploration and exploitation through the combined efforts of population- and trajectory-based algorithms. In the third phase, the search is confined to the most promising sub-region identified earlier. The trajectory-based algorithm then leverages the best solution’s fitness value and position to effectively explore this restricted search space. We compare the proposed approach with various metaheuristics in ten well-established unimodal and multimodal continuous optimization benchmarks. The results highlight the ability of the new hybrid approach to identify global optima while reducing execution time.
In this paper we expand on a previously presented method to autonomously create a reduced model to describe aspects of a computational fluid dynamics (CFD) simulation. We are trying to reduce the particle tracking of the CFD simulation so that we can determine the focal points of each particle simply by its initial injection parameters. In order to do so autonomously we utilize a Gaussian process (GP) which has the main advantages of also being able to depict random effects through a confidence of the model. While we exemplify this method through the case of the particle tracks in our Aerosol-on-Demand (AoD) jet-printhead, we see the CFD model as the ground truth for the purposes of validation. Our goal is to have the reduced model be as close as possible to the computationally expensive CFD simulation so that the reduced model can be used inline in future use and rapid iteration to improve the printing process.
The optimization of job-shop scheduling problems, such as those found in the semiconductor industry, is an NP-hard challenge. Research has demonstrated that agent-based modeling of production plants can effectively plan tasks, maximize productivity (utilization and timeliness), and minimize production delays. This bottom-up optimization approach especially addresses the computational challenges associated with traditional, centrally calculated optimization methods. In this context, we focus on a dynamic semiconductor production plant, modeling both machines and products as agents. We propose two variants of the Artificial Bee Colony algorithm for scheduling from the bottom up. Variant (1) emphasizes decentralization and batch processing to enhance production speed, while Variant (2) aims to predict production times to reduce queuing delays, achieving reductions in Flow Factor and Tardiness across most lots, though with a slight increase in Makespan for a few cases. Both algorithmic variants are evaluated within the SwarmFabSim framework, which is designed in NetLogo, specifically addressing the job-shop scheduling problem in the semiconductor industry. Through this study, we analyze the effectiveness of the bottom-up algorithms that rely on low-effort local calculations.
