AI Planner of the Future
The AI (Artificial Intelligence) PLANNER OF THE FUTURE ambitious research program focuses on the increasing intertwining of technology and human aspects in the context of AI planning for supply chains and logistics. The European Supply Chain Forum (ESCF), the department Industrial Engineering and Innovation Sciences (IE&IS), the Eindhoven Artificial Intelligence Systems Institute (EAISI), and the Logistics Community Brabant signed a long-term collaboration agreement.
Reach out to us to be part of this ambitious research program with your data, challenges, and all AI planning-related questions!
The AI planner of the future program needs a broad range of research fields and a rich set of involved companies. We have both! We combine the extensive knowledge of researchers from all multi-disciplinary IE&IS domains and the real-life living labs of European Supply Chain Forum companies from diverse industries. All industries are involved: fast-moving consumer goods, omnichannel retailing, last-mile logistics, services, health, transport and mobility, high-tech industries, etc.
The research program is a unique research and valorization network. It combines 25 Artificial Intelligence researchers, 10 Ph.D. students, and over 50 Bachelor and Master students, for five years (2021-2026). This program is hosted by the TU/e-based Department of Industrial Engineering & Innovation Sciences. It is supported by the European Supply Chain Forum, the Department of Industrial Engineering & Innovation Sciences, the Eindhoven Artificial Intelligence Systems Institute (EAISI), and Logistics Community Brabant.
There is more to AI in logistics than building algorithms to predict an outcome. It is about extracting the right data, understanding how people make decisions with the help of algorithms, and deriving long-term managerial insights. The strength of our program lies in studying the people-AI-logistics interface with an interdisciplinary team.
The transition towards a more data-driven supply chain and a more sustainable one should go hand in hand.
In research and industry contexts, we need to keep in mind how innovations around AI relate to digital society and sustainability challenges
Learning about Customers: Demand Implications of Logistics-Related Decision-Making in B2B
In business-to-business (B2B) exchanges, customers are more likely to buy from suppliers who know them well and consistently provide good service. Yet, when planners optimize the operations with a focus on cost-reduction, they risk overlooking the importance of building long-term relationships with their customers. These relationships critically depend on learning the customer’s preferences, priorities, and service expectations. While building strong customer relationships in a B2B context has traditionally been the salespeople’s responsibility, AI developments now open the possibility for AI-based learning about customers. AI-based learning about customers in a B2B setting is complex though because each customer has their own needs and preferences, leading to highly customized offerings. These customized offerings often include agreements on critical logistics-related decisions such as lead times, delivery, and maintenance planning. In this setting, close contact between the people from sales and operations – i.e., a strong marketing-operations interface – benefits the customer relationship. Yet as information on customers is embedded both in IT systems (e.g., CRM systems) and people (e.g., salespersons), this is a domain where B2B firms can benefit greatly from AI. This PhD project thus studies on how AI can help planners tailor their operations to better serve customer needs.
Context matters: optimizing shared decision making in real-world forecasting and inventory management
In many organizations and across industries, artificial intelligence (AI) is transforming the way we work. AI-systems are implemented to assist employees with decision making, to decrease workload, or to increase efficiency. Although promising, transforming traditional operations into ones that rely on autonomous systems brings many challenges. For example, when using AI planning systems, users frequently experience difficulties in using and trusting these systems and, as a consequence, deviate from their advice. Prior research highlights the impact of system characteristics (e.g. reliability) on human-AI collaboration. However, these studies disregard the important influence of contextual factors on human-AI collaboration. Therefore, one important challenge concerns the consideration of contextual factors when designing and implementing AI-systems at work. To successfully integrate these systems in organizational processes, it is critical to understand when and why users are (un)willing to adopt these systems in their work routines and how we can stimulate effective usage. The goal of this PhD-project is to address these issues by answering the following research questions: (1) Which contextual factors, specifically organizational (e.g. organizational climate, leadership) and societal factors (e.g. COVID-19 pandemic), impact planners’ willingness to use AI planning systems? (2) How can human-centered AI and work design help to improve human-AI collaboration?
AI-Based Replenishment and Order Fulfillment Strategies for Omnichannel Supply Chains
While retailers are investing heavily in integrating online channels to their traditional offline channels, only a few giant players’ efforts are profitable. The main problem is the failure to integrate essential operations of online and offline channels. To tackle this problem, this PhD research provides a real-time, data-driven AI-based planning strategy by integrating the decisions on three cornerstones of omnichannel retail: 1) inventory replenishment policy, 2) customer fulfilment policy, and 3) consumer delivery options and prices. We provide a new AI-based methodology to enable the real-time integrated control of these three cornerstones. Our study provides the omnichannel retailers with decision support tools that tell them when and where products in the supply chain should be replenished and stored and how customer orders should be fulfilled. In this way, we identify profitable omnichannel business models that will help these businesses to stay alive in the e-commerce market.
Robust data-driven sustainable cold chain
Cold chains contain different types of uncertainty; from quality of supplied material to traffic congestion in last-mile delivery. These uncertainties often result in product losses and waste, aggravating the environmental footprint of the chain. As such, there is a need for robust policies that can address this uncertainty. Recently, we have seen how machine learning techniques, such as Neural or Kernel-based classifications, are used in Robust Optimization to derive robust policies by extracting important information from historical data. However, the computational complexity of these approaches still remains an issue. In this project, we want to build on this recent stream of research and design approaches in data-driven robust optimization capable of tackling problems arising in cold chain settings. By using machine learning techniques within the framework of robust optimization we intend to then design robust policies that mitigate the negative effects of uncertainties that can be found within real-world settings such as the food system or pharmaceutical supply chains.
Digital Twins: An ingenious AI companion or an evil twin?
Digital twinning and remote visualization technologies rapidly gain popularity for the design and maintenance of (complex) production systems. Digital twinning is not a new term but paired with advancements in artificial intelligence (AI) and augmented reality (AR), it is increasingly valuable in transforming industrial operations, which, in turn, leads to the creation of additional business value. Digital twinning involves embedding sensors in Internet-of-Things-connected, complex industrial machines and applying artificial intelligence and machine-learning algorithms to the resultant big data. A sophisticated visualization of the machine allows remote engineers to proactively optimize productivity, reduce maintenance cost, and extend product life cycles. Although proactive actions make the manufacturer seem “closer” to the customer’s business than ever before, paradoxically, the remote elements in digital twinning limit real-life customer contact that is needed to build loyal customer relationships and to gather ideas for new and improved products. Hence, while digital twinning can be an ingenious companion in optimizing operational decisions, it may also act as an evil twin that hampers marketing and innovation outcomes. Manufacturers need a solution for this pressing issue, but current literature has not yet considered the potential dark side of digital twinning in an interdisciplinary manner.
AI for sustainable last-mile delivery by micromobility: a socio-technical perspective
AI and micromobility are both considered as promising solutions for increasing the sustainability of last-mile delivery. While AI solutions may improve the efficiency of last mile logistics, only a shift to clean mobility modes holds the promise of zero emissions. Micromobility such as transport by e-cargo bikes is such a mode. Last mile logistics using e cargo-bikes (ideally with associated micro hubs) can be low cost, flexible and more distributed than traditional delivery modes. Realising the potential of AI logistics for micromobility does however crucially depend on the availability of suitable data (e.g. training sets for ML), while most data practices and AI approaches occur around traditional modes such as vans. This project therefore conducts a socio-technical analysis of how the intersection of two innovations – micromobility and AI-driven last-mile delivery – could lead to more sustainable urban logistics. It identifies opportunities for synergies, but also critically interrogates claims and practices around AI and sustainability. Active micromobility such as cargo-ebikes will be a particular focus.
Data-driven Optimization using Digital Twins for Sustainable Last-Mile Delivery
Due to the complexity of modern supply chains, it is difficult to predict what the effect will be of a decision aimed at reducing greenhouse gas emissions, such as choosing the location of a pick-up point, or changing the travel route for a vehicle. Digital twins make it possible to try these decisions in a virtual environment before applying them in real life. This helps policymakers in governments and companies gain a better understanding of the consequences of a decision, which reduces the risks and uncertainties of the radical new decisions that are necessary to achieve the sustainable supply chain of the future. With the rise of digital twins for smart cities, such as the Atlas Livable City developed by the Logistics Community Brabant, more data is readily available than ever before. Yet most existing optimization techniques, which are necessary for minimizing an objective such as travel time or greenhouse gas emissions, are not able to deal with such complex virtual environments. Data-driven optimization techniques are therefore an active area of research. Examples of this are optimization heuristics learned with machine learning, and surrogate models for optimization. This project will contribute to this active research landscape by making data-driven optimization techniques that are suitable for digital twins. The main application is the reduction of greenhouse gas emissions in last-mile delivery by choosing the locations of pick-up points in urban environments.
Online Supply Chain Planning
Project Supply chain planning is complex because of complex dependencies of the delivery of the final product on the timely shipping, assembly, production, and procurement of its parts. Periodic supply chain plans are made. However, during the execution of the plan, it must often be adapted, for example, because parts are delivered late or because production is delayed. This leads to changes to the plan that are often ad-hoc and suboptimal, and cause planning ‘nervousness’, i.e. frequent planning changes. Consequently, in addition to periodic planning, supply chain planning can benefit from planning techniques that assist with the day-to-day adaptations of the supply chain plan, due to the unexpected situations that arise. These ‘online’ planning techniques must take the current periodic plan into account, as well as the current status of the procurement, production, and assembly of the parts. It should then advise on changes to the plan, while minimizing planning nervousness and costs. In this project we aim to develop such a technique for online supply chain planning, using novel techniques from the area of artificial intelligence that can learn to predict – based on the current situation and unexpected events that must be handled – what the best solution is to plan for the unexpected event. A general framework for these techniques is being developed in a related project, where applications in production and transportation planning are studied. The aim of this project is to make this general framework suitable for supply chain planning. Against this background, the project has a specific focus on encoding and learning the complex relations and patterns of dependency between different activities in supply chain planning, which to the best of our knowledge has not been studied before.
From feared competitor to trusted companion: understanding and enhancing trust in AI over time
Artificially Intelligent systems are becoming both much more pervasive, and better. There is a lot of evidence though that the interaction between the human planner and AI systems is far from hassle free: AI-generated decisions are overridden or adapted when they should have been left alone, and AI-systems are trusted when they should not have been. The literature has suggested several factors that influence the trust that a planner has in AI systems, some related to the planner (experience and expertise, for instance), some related to the system (transparency, reliability, fairness, …) and some related to the context in which the interaction takes place (high-risk vs low-risk decisions, complex vs more standard, …). An overlooked issue is that in many organizations planners interact with the AI system repeatedly. This causes that, as planners interact more often, how they feel about and behave towards the system becomes more and more dependent on their experience with the system (and less dependent on these more often studied initial factors). This project focuses on trust in AI-systems over time and how past interactions of the planner with the AI-system shape future interactions.
Widening the frame: Rational choice beyond a given utility function
Supply chain and logistics planning problems can be seen as optimisation problems that require collecting as much relevant information as possible, determining possible choices, and selecting the action with the highest expected utility. They thus lend themselves to AI solutions that use the same model: “… we build optimising machines, we feed objectives into them, and off they go.” (Russell 2019, 172). “Rational choice” in this sense assumes a given utility function. But apart from well-known problems with rational choice in real-world environments (e.g. uncertainty, dynamic changes, other agents, non-discreteness of actions), we know from the human example that highly complex choices in real-world environments require metacognition, e.g. considering which utility function to use, whether our reasoning is trustworthy, whether knowledge is sufficient, whether to act now or to optimise the decision further, whether a course of action is ethical. Humans (and certain animals) are able to change the frame of reference and move to metacognition, when needed. The supply chain and logistics planning problems are a fine place for a case study of this metacognition problem in a practical environment. When and how should a system say: “It is best not decide this and act now, I should change the frame”?