CHAIR OF

MANAGEMENT

SCIENCE

Pro-Active Routing for Emergency Testing in Pandemics, funded by DFG since 2022

A pandemic can immobilize municipalities within a short amount of time. The key is to discover and avoid spreading of infection clusters through fast and effective testing. An innovative idea implemented during the COVID-19 pandemic in metropolitan areas such as Vienna, Austria, is the employment of a workforce of mobile testers. This project deals with the operational management of such mobile testers and the resulting impact on the spread of a disease using COVID-19 as an example. Based on state-of-the-art multi-agent simulation models, we will generate and analyze data on the temporal and spatial spreading (descriptive analytics). With methods of predictive analytics, we will aggregate the data to a detailed information model with a particular focus on modelling correlation for testing demand. Using this, we will model and solve the dynamic tester routing with infection hot spots and correlation demand problem (TRISC) using methods of prescriptive analytics, esp. reinforcement learning. The obtained policies will be evaluated by the multi-agent simulation again. The following core research questions will be investigated: (1) How can data of the spread of highly infectious diseases like COVID-19 be analyzed and modeled for the purpose of dynamic workforce control? (2) How can we achieve an effective dynamic control of the workforce in reaction and in anticipation of the complex disease information? (3) When is anticipatory dynamic workforce control effective in containing the spread of pandemics? The problem at hand shows new and severe complexity in the information model of the demand (test requests) and in the decision model for the operational control. Deriving the demand information model (via predictive analytics) is complex because it must capture the spatial-temporal correlation of demand. The decision model for the problem is a novel stochastic and dynamic vehicle routing problem. Determining high-quality decisions that integrate the information model (via prescriptive analytics) is therefore additionally challenging. The evaluation by an established agent-based simulation is particularly exceptional for this research field. The project will be conducted by Jan Fabian Ehmke (JE, Universität Wien), Marlin Ulmer (MU, Technische Universität Braunschweig), and Niki Popper (NP, Technische Universität Wien). JE will serve as coordinator and is responsible for tasks of predictive analytics. MU leads the project part on prescriptive analytics for dynamic vehicle routing. NP will contribute with an agent-based simulation that supports the creation of the predictive information model and the evaluation of dynamic and stochastic disease sampling. This will provide unique opportunities to extend current methods including their evaluation in the urgent application of disease routing.

Please find information about the project on the website of DFG here.

Integrating machine learning in combinatorial dynamic optimization for urban transportation services, funded by DFG since 2022

The goal of this project is to provide effective decision support for stochastic dynamic pickup and delivery problems by combining the strengths of mixed-integer linear programming (MILP) and reinforcement learning (RL).Stochastic dynamic pickup-and-delivery problems play an increasingly important role in urban logistics. They are characterized by the often time-critical transport of wares or passengers in the city. Common examples are same-day delivery, ridesharing, and restaurant meal delivery. The mentioned problems have in common that a sequence of decision problems with future uncertainty must be solved in every decision step where the full value of a decision reveals only later in the service horizon. Searching the combinatorial decision space of the subproblems for efficient and feasible tours is a complex task of solving a MILP. This complexity is now multiplied by the challenge of evaluating such decision with respect to their effectiveness given future dynamism and uncertainty; an ideal case for RL. Both are crucial to fully meet operational requirements. Therefore, a direct combination of both methods is needed. Yet, a seamless integration has not been established due to different reasons and is the aim of this research project. We suggest using RL to manipulate the MILP itself to derive not only efficient but also effective decisions. This manipulation may change the objective function or the constraints. Incentive or penalty terms can be added to the objective function to enforce or prohibit the selection of certain decisions. Alternatively, the constraints may be adapted to reserve fleet-resources.The challenge is to decide where and how the manipulation takes place. SDPDPs have constraints with respect to routing, vehicle capacities, or time windows. Some constraints may be irrelevant for the fleet’s flexibility while others might be binding. The first part of the research project focuses on identifying the “interesting” parts of the MILP via (un-)supervised learning. Once the “interesting” parts are identified, the second challenge is to find the right parametrization. Here, we will apply RL methods to learn the state-dependent manipulation of the MILP components.

Please find information about the project on the website of DFG here.

Urban Mobility and Logistics: Learning and Optimization under Uncertainty, funded by DFG since 2020

The goal of this project is to systematically improve quantitative decision support for urban mobility and logistics, to analyze its methodological functionality, to derive general conceptual insight, and to use the derived concept for future method designs.For applications in urban mobility and logistics, operational decision support needs to be effective, fast, and applicable on a large scale – often under incomplete information. Providers face uncertainty in many components, for example, the customer demand, the urban traffic conditions, or even the driver behavior. Mere adaptions to new information are often insufficient and anticipation of this uncertainty is key for successful operations. In research and practice, a range of anticipatory methods has been developed, usually tailored to specific practical problems. Such methods may follow intuitive rule-of-thumbs, draw on sampling procedures, or use reinforcement learning techniques. While the methods may perform well for individual problems, there is still a very limited understanding of the general dependencies of a method’s performance and a problem’s characteristics. This research project will provide this conceptual understanding.To this end, the project will systematically develop and compare different methodology for a set of problems from three different application areas, one combining urban mobility and transport as a service, one using a network of parcel stations for urban transportation, and one performing pickup and delivery with a gig economy workforce. The three problems differ in several dimensions, especially in their sources of uncertainty. To classify the problems, measures will be developed, for example, with respect to the scale of the problem or structure and degree of uncertainty. For each problem, a set of different methods will be developed. The methods will improve decision support for the specific problems while simultaneously allowing a systematic analysis of dependencies between problem and methodology performance. To this end, additional measures will be developed to classify method performance, for example, decision speed, or the interpretability of a method. Based on the problem and method measures and the extensive experiments and analyses, a framework will be developed to guide future method design for this emerging research field.

Please find information about the project on the website of DFG here.

Combined Approximate Dynamic Programming for Dynamic Same-Day Delivery, funded by DFG from 2018 to 2022

E-Commerce has increased sales by two-digit percentages in the last years. In the future, same-day delivery (SDD) will become a major success factor for E-Commerce companies. However, offering SDD is expensive because short delivery deadlines and subsequently ordering customers leave little room for consolidation. To cost-efficiently provide SDD, decision support methods are required. On the operational level, these methods dynamically create, update, and adapt delivery tours based on newly revealed information. For effective decision making, these methods need to anticipate both the detailed short term impact as well as the general long-term impact of a decision. SDD-problems form a subgroup of stochastic dynamic vehicle routing problems. This problem class is relatively new and general methods are not established yet. Because of the high complexity of dynamic vehicle routing problems, exact methods cannot be applied. First work in this area draws on heuristic methods of approximate dynamic programming (ADP). ADP-methods use simulation of the dynamic model to approximate a decision’s impact on the future. These methods can be differentiated based on the time these simulations take place. Online methods start simulating in the actual decision state. Offline methods conduct simulations before the decision process starts. They store the aggregated results and access them during the actual decision process. Online methods can simulate using full detail of a decision state but only with limited calculation time available. Offline methods allow frequent simulations and reliable long-term approximations, however, on an aggregated level. For the SDD-problem at hand, both short-term detail and long-term reliability are essential for successful decision support. However, both online and offline methods fall short in one of the two capacities. A combination is necessary. This research projects aims on developing a combined ADP-method for the SDD-problem. The method allows a generic, state-dependent shift between online and offline simulation results. The method will provide effective decision support and business insight for a new and important SDD-problem. Further, this method will be generic and broadly applicable in the field of dynamic vehicle routing. It will therefore be an important step towards a general solution framework in dynamic vehicle routing.

Please find information about the project on the website of DFG here.