Alexander Klaas

A new approach for analyzing convergence algorithms for mobile robots

Given a set of n mobile robots in the d-dimensional Euclidean space, the goal is to let them converge to a single not predefined point. The challenge is that the robots are limited in their capabilities. Robots can, upon activation, compute the positions of all other robots using an individual affine coordinate system. The robots are indistinguishable, oblivious and may have different affine coordinate systems. A very general discrete time model assumes that robots are activated in arbitrary order. Further, the computation of a new target point may happen much earlier than the movement, so that the movement is based on outdated information about other robots positions. Time is measured as the number of rounds, where a round ends as soon as each robot has moved at least once. In [6], the Center of Gravity is considered as target function, convergence was proven, and the number of rounds needed for halving the diameter of the convex hull of the robots positions was shown to be O(n2) and Ω(n). We present an easy-to-check property of target functions that guarantee convergence and yields upper time bounds. This property intuitively says that when a robot computes a new target point, this point is significantly within the current axes aligned minimal box containing all robots. This property holds, e.g., for the above-mentioned target function, and improves the above O(n2) to an asymptotically optimal O(n) upper bound. Our technique also yields a constant time bound for a target function that requires all robots having identical coordinate axes.

Collisionless gathering of robots with an extent

Gathering n mobile robots in one single point in the Euclidean plane is a widely studied problem from the area of robot formation problems. Classically, the robots are assumed to have no physical extent, and they are able to share a position with other robots. We drop these assumptions and investigate a similar problem for robots with (a spherical) extent: the goal is to gather the robots as close together as possible. More exactly, we want the robots to form a sphere with minimum radius around a predefined point. We propose an algorithm for this problem which synchronously moves the robots towards the center of the sphere unless they block each other. In this case, if possible, the robots spin around the center of the sphere. We analyze this algorithm experimentally in the plane. If R is the distance of the farthest robot to the center of the sphere, the simulations indicate a runtime which is linear in n and R. Additionally, we prove a theoretic upper bound for the runtime of O(nR) for a discrete version of the problem. Simulations also suggest a runtime of O(n + R) for the discrete version.

Fast converging, automated experiment runs for material flow simulations using distributed computing and combined metaheuristics

The analysis of production systems using discrete, event-based simulation is wide spread and generally accepted as a decision support technology. It aims either at the comparison of competitive system designs or the identification of a “best possible” parameter configuration of a simulation model. Here, combinatorial techniques of simulation and optimization methods support the user in finding optimal solutions, but typically result in long computation times, which often prohibits a practical application in industry. This paper presents a fast converging procedure as a combination of heuristic approaches, namely Particle Swarm Optimization and Genetic Algorithm, within a material flow simulation to close this gap. Our integrated implementation allows automated, distributed simulation runs for practical, complex production systems. First results show the proof of concept with a reference model and demonstrate the benefits of combinatorial and parallel processing.

Simulation aided, knowledge based routing for AGVs in a distribution warehouse

Traditional routing algorithms for real world AGV systems in warehouses compute static paths, which can only be adjusted to a limited degree in the event of unplanned disturbances. In our approach, we aim for a higher reactivity in such events and plan small steps of a path incrementally. The current traffic situation and also up to date time constraints for each AGV can then be considered. We compute each step in real time based on empirical data stored in a knowledge base. It contains information covering a broad temporal horizon of the system to prevent costly decisions that may occur when only considering short term consequences. The knowledge is gathered through machine learning from the results of multiple experiments in a discrete event simulation during preprocessing. We implemented and experimentally evaluated the algorithm in a test scenario and achieve a natural robustness against delays and failures.

Using simulation as an adaptive source of knowledge for the control of material handling systems

Recent advances in computing have allowed simulation to be used as a source of data in the real-time control of logistics systems such as Material Handling Systems (MHS). For a real-world MHS, in development by Lödige Industries GmbH, Germany, we demonstrate the benefit of generating data offline using a parametrized simulation model that real-time operational control is based on. The data consist of mappings of control situations to optimal actions respectively. Our approach allows for self-adaptation of the simulation by observing current system parameters that are fed into the model. The control automatically triggers regeneration when necessary, detects changes in the system and also proactively anticipates them, resulting in consistently high performance. We furthermore use a simulation-based look-ahead method to consider uncertainties when evaluating alternative actions. Evaluation results show a significant increase in system performance compared to fixed application of a control action and demonstrate the benefits of the self-adaptive properties.

Simulation aided, self-adapting knowledge based control of material handling systems

Knowledge based methods have recently been applied to the control of material handling systems, specifically using simulation as a source of knowledge. Little research has been done however on ensuring a consistently high quality of the data generated by the simulation, especially under changing circumstances such as differing load patterns in the system. We propose a self-adapting control that is able to automatically generate knowledge according to current circumstances using a parametrized simulation model, which uses observed system parameters as input. The control automatically triggers generation when necessary, detects changes in the system and also proactively anticipates them, resulting in consistently high performance. For the problem of knowledge generation (determining an optimal control action to a given situation), we present a look ahead simulation method that considers uncertainties. We validated our approach in a real world material handling system, developed by Lödige Industries GmbH.

Visualization and Collaborative Editing of Simulation Models With Heterogeneous Clients: Implemented Into the Simulator D3FACT

Today’s simulation software normally has fixed, built-in editing and visualization views adapted to a specific problem domain. Often this monolithic concept prevents collaborative editing altogether. Even with more flexible concepts, editing simulation models with a heterogeneous set of clients is not possible. In this article we describe a flexible concept for collaborative editing of simulation models with heterogeneous clients, such as web-based, desktop and mobile clients. The clients may even show different editing views adapted to the user’s role. The concept we describe in this paper overcomes several problems: First we need to be able to connect and manage a set of heterogeneous clients in the simulation software. The very different user inputs from the connected clients then need to be processed, interpreted and combined to allow editing in a collaborative way for all users. At last we show a prototypical integration of the presented concept into our research platform d3fact.Copyright

Decentralized Real-Time Control Algorithms for an AGV System

Automated guided vehicles (AGVs) are increasingly being used to transport goods or people. Navigation is a core issue in such a system. The AGVs use real-time control algorithms to reach their assigned destination autonomously. For reasons like scalability and flexibility, it is beneficial that the shuttles compute the necessary calculations decentrally. In this paper, we present such decentralized algorithms for conflict-free routing in a specific AGV system. Based on existing algorithms for deadlock handling in theory and routing in computer networks, we implemented three different sets of algorithms of varying sophistication in a logistics simulator. Evaluation reveals their functionality and relative performance.