How Augmented Reality could transform last-mile logistics as we know

1

How Augmented Reality could transform last-mile

logistics as we know

University of Groningen

Faculty of Business and Economics

MSc Supply Chain Management

MSc Technology and Operations Management

Master Thesis

Student: Jelmer Winkel

Student number: S3524914

E-mail: j.h.winkel@student.rug.nl

Supervisor: dr. ir. P. Buijs

Second Accessor: dr. ir. S. Fazi

2

Abstract

The rapid growth of e-commerce is challenging parcel delivery companies to meet the demand while keeping the costs down. On average, the last-mile is accountable for 28% of the total logistical costs. Different solutions are proposed drive those costs down by increasing efficiency. This thesis discusses one of those state-of-the-art solutions: Augmented Reality (AR). AR has the potential to project relevant information to the drivers in the last-mile to increase their performance. This thesis presents the first empirical study that investigates the effect of AR on the efficiency in last-mile logistics. A single case study was conducted with an innovative company and a parcel delivery company. The thesis presents an AR-system design as well as test results from an lab experiment verifying its functionalities. The outcome of the experiment showed that AR could be used in last-mile logistics. The effect on the efficiency was validated through a computational study that would compare different scenarios with the current situation. The results show that substantial cost savings can be achieved when using AR. The technology is able to increase the speed of delivery and reduce training costs. The most important finding was that the AR-system has the ability to decouple loading from the last-mile logistics and generate a significant improvement in vehicle utilization As a result, parcel delivery companies can reduce their overall costs or can anticipate the continuing growth in demand.

Keywords: last-mile logistics, Augmented Reality, AR, parcel delivery companies,

3

Table of Contents

3.2.1 Virtual lab experiment... 15

3.2.2 Simulation of the last-mile logistics ... 16

4 Findings ... 18

4.1 AR-system design ... 18

4.2 Tests ... 20

4.2.1 Tracking the location of the parcel ... 20

4.2.2 Projecting the location ... 23

5 Computational study ... 25

5.1 Computational design ... 25

5.2 Computational Study ... 28

4

6.1 Theoretical implications ... 32

6.1.2 Last-mile logistics ... 33

6.2 Managerial implications ... 33

6.3 Limitations and Future research ... 34

5

1 Introduction

The last-mile logistics is challenged by the rapid growth in e-commerce (Wang et al, 2016). While it forms only the very last stage of the supply chain, and the parcels are transported over the shortest distance, it is accountable for 28% percent of the total logistical costs (Gevaers et al., 2009; Wang et al., 2016). Most research that has been conducted on increasing the efficiency of last-mile logistics focused on sorting and route-optimization (Allen et al., 2007; Nathanail et al., 2016). Other labor-intensive activities, such as loading and final delivery, have been given far less attention. The rise of new technologies, such as Augmented Reality (AR), can create new opportunities to improve the efficiency of these labor-intensive activities.

AR can display specific information as a virtual layer on top of the real world (Azuma, 1997) and is seen by many as a technique that could be used to improve the efficiency of the logistics (Glockner, Jannek, Mahn, 2014; Reif et al., 2010; Van Krevelen & Poelman, 2010). Several authors have investigated the application of AR in other areas of logistics research, such as in warehousing operations (Reif et al., 2010; Rejeb, 2019). Field tests have shown that AR can reduce lead times for order pickers in warehouses compared to searching with a paper picking list (Reif et al., 2010). AR directly provides the information to the order picker instead of having to search for the next order on the paper list (Glockner, Jannek, Mahn, 2014; Van Krevelen & Poelman, 2010). Others, however, doubt the actual benefits that AR can bring because of several challenges during the implementation, such as hardware or software limitations (Rejeb, 2019; Stoltz et al., 2017). These authors claim that there is a lack of empirical data to support the success of AR in logistics. This thesis aims to contribute to this continuing academic debate by extending AR to the last-mile logistics and provide empirical data to explain its potential.

6 driver’s field of view (Ong et al., 2008; Reif et al., 2010). Lastly, AR could be used to decouple loading which is a labor-intensive process that takes up a significant amount of time (Glockner, Jannek, Mahn, 2014) – indeed, valuable time that could also be used to deliver extra parcels. Thus, if AR is capable of performing one of these functionalities it can provide the opportunity to deliver more parcels and consequently increasing the efficiency of last-mile logistics. So far, no research has been conducted to study the effect of AR on the performance of last-mile logistics and this has led to the following research question.

1. How can AR be used to increase the efficiency of last-mile logistics? 1.1 How can AR be designed for application in the last-mile logistics? 1.2 What efficiency gains can be expected from using AR in the last-mile?

A single case study was conducted with multiple perspectives from a leading innovating company and an established parcel delivery company. Together with the innovating company, a AR-system design for the last-mile logistics was developed. A virtual lab experiment empirically verified the system’s capabilities. Interviews and observations at the parcel delivery company provided input for developing a mathematical model to quantitatively validate the benefits that AR can have on the efficiency of the last-mile logistics processes.

The contribution of this thesis is threefold. First, this paper is one of the first to provide empirical evidence to support the academic discussion on the effectiveness of AR. Second, this paper proposes a system design that changes the definition of last-mile logistics as we know it. Third, this research provides insights on the effect of AR on the different processes to improve mile efficiency. This opens doors for new debates on the effectiveness of the current last-mile logistics processes and whether it would be more favorable to reshape it.

7

2 Theoretical Background

2.1 Augmented Reality

Augmented Reality (AR) combines the physical world with a virtual world and can be used to provide information via a virtual layer displayed to the user. Since its inception, AR is seen as an exciting technology that can be applied in different industries such as medical, manufacturing, entertainment or the military (Azuma, 1997). AR can be seen as the middle ground between the virtual world (Virtual Environment) and the real world (Real Environment) (Milgram et al., 1995). In Figure 1, the reality-virtuality spectrum is depicted. The objective with AR is to add information to a real object or place (Azuma, 1997; Cai et al., 2011; Milgram et al., 1995). This information can enhance a user’s understanding of the environment they are in. AR is defined by 3 key characteristics for AR namely: (1) it combines the real and virtual (2) is interactive in real-time and (3) is registered in three dimensions (Azuma, 1997). If the technology does not check each of the boxes, it cannot be considered as AR.

Figure 1 - Reality-Virtuality spectrum.

2.1.1 The technology

8

2.1.2 AR system landscape/architecture

AR is not a stand-alone technology but instead fits in a more extensive system. Over the last years, several systems designs and architectures have been created to support the use of AR. Essential elements of those architectures are a smart device (smart-glasses or mobile phone), a projecting application system, algorithms, camera vision, internet connection, privacy gates, command controls, modeling tools, backend services and databases (Zobel et al., 2018). The system architecture is used as a reference model to create the AR-system for last-mile logistics and is presented in Figure 1.

Figure 1 - Reference architecture for smart glasses-based systems (Zobel et al., 2018)

9 The BACKEND modules consist of software programs and services that enable the system to work such as backend services, client services, algorithms and databases. The algorithms are essential for recognizing the environment and accurately track and project the location of the objects. The accuracy of the algorithms is optimized through machine learning and deep learning. Deep-learning is an evolved form of machine learning and represents the human way of learning through example/experience (Lampropoulos et al., 2020). Deep learning is an ideal technique and is widely adopted in the computer vision domain because it is effective for solving complex problems such as object classification and recognition (Lampropoulos et al., 2020). The camera vision will classify each parcel with a unique visual appearance such as name, address, barcode, and physical appearance. The algorithms will use these attributes to be able to differentiate parcels. The smart device is connected through a cloud to the BACKEND system and enables real-time data. This will allow a system to make use of cloud computing, which is the ability to process information remotely, and thus decrease the technical requirements from the smart devices (Ding et al., 2020). Multiple papers have addressed computing power as one of the technical challenges for adopting an AR system (Rejeb, 2019; Stoltz et al., 2017). Solving this issue through cloud-computing will bring the adoption one step closer to realization. However, it could be argued whether the constant search for a cellular network will not drain the battery life more than computing locally.

2.1.3 Benefits

10 regarding AR in logistics has solely focused on its effectiveness on activities in the warehouse (Glockner, Jannek, Mahn, 2014; Reif et al., 2010; Rejeb, 2019; Stoltz et al., 2017). These studies have shown that AR leads to a decrease in error-rates for picking and increased speed for sorting and routing.

2.1.4 Challenges

There are also challenges and barriers when adopting AR. These challenges can be divided into technical, organizational, economic, and ergonomic challenges (Rejeb, 2019; Stoltz et al., 2017; Van Krevelen & Poelman, 2010). Rejeb (2019) conducted a systematic literature review of 43 papers from academic journals on the challenges of AR in logistics. From these papers, 74% mentioned technical challenges as an obstacle to the adoption of AR systems. The other organizational and ergonomic challenges are a consequence of the immaturity level of technology. Technical challenges are, for instance, related to hardware such as battery life, wearables, or smartphone cameras. The wearables should be able to last for an entire workday and display high-quality images and that requires a powerful battery and high-quality video cameras. A key challenge is object recognition for AR-systems because the virtual and real objects have to be in harmony to achieve high accuracy (Uma, 2019). Developments on the robustness of the system are still improving for accurate real-time tracking and computational costs are high (Van Krevelen & Poelman, 2010). Furthermore, software challenges regard no standardization of the programming language and this makes it difficult for practitioners to experiment and create the correct algorithm (Stoltz et al., 2017). Economic challenges are the high costs of ownership and the high initial investment required when introducing a new type of information technology.

2.2 Last-mile logistics

11

Figure 2 – Last-mile activities

2.2.1 Loading

Loading is an essential part of the last-mile delivery process because it is a labor-intensive and time-consuming activity parcel delivery company (Bartholdi & Gue, 2000; Fazili et al., 2017). Drivers load their truck using a common technique in the parcel delivery industry, namely, the Last-In-First-Out (LIFO) technique (Moura & Oliveira, 2009; Veenstra et al., 2017). This technique means that the first parcel that needs to be delivered is the last parcel that is loaded into the truck. Therefore, the driver needs to check the planned route, scan the parcels and load the truck accordingly. LIFO provides the driver with a structured approach for finding the parcel when performing the delivery and increases the chance that he or she remembers where the parcel is located. Additionally, LIFO prevents the unnecessary moving of parcels because the parcel is located in the back and thus avoids additional handling costs when performing the delivery (Liu et al., 2019; Veenstra et al., 2017). Drivers spend around 40% of their time at the warehouse and one significant contributor is the loading process (Glockner, Jannek, Mahn, 2014). In case that loading time can be reduced, the drivers are longer on the road and can deliver more parcels.

2.2.2 Transport

12 a mechanism to avoid traffic jams and increase the efficiency of last-mile (De Marco et al., 2018). Another way to increase efficiency is to optimize the route of the vehicle. This is by far the most studied subject in the last-mile research (Demir et al., 2014; C. Lin et al., 2014; Purnamasari & Santoso, 2018; Tsai et al., 2003) and has resulted in considerable improvements in routing efficiency. The effect that VRP has on the total costs is estimated with a reduction of 5% to 20% (Cattaruzza et al., 2017). However, more elements of transport could have the potential to be improved. For instance, for finding a parking spot, parking it-self or avoiding traffic jams (Fransoo, 2019; Glockner, Jannek, Mahn, 2014).

2.2.3 Delivery

Home deliveries will continue to take place, and therefore, it is interesting to look into methods to increase the efficiency of it. The delivery forms the only moment of contact between the parcel delivery company and the customer. The delivery process consists of activities related to searching the parcel, scanning it, delivering it to the consumer, and signing-off (Glockner, Jannek, Mahn, 2014). The service-level can measure the quality of the delivery. The service level can, for instance, be measured by the deviation from the proposed delivery time window and the actual delivery time. A smaller time window will increase the service-level because the customer has to stay at home for a shorter period of time (Punakivi et al., 2001). Therefore, parcel delivery companies must have an accurate capacity planning to ensure they deliver in the proposed time window. A comparative analysis of last-mile delivery systems performed by Allen (2007), showed that the attended delivery realizes the highest delivery costs compared to locker-banks or collection points. See Appendix A for a full overview of the analysis. The reason for these high delivery costs are the high no. of failed deliveries and the high drop-off time. The failed deliveries are usually the result of the customer not being at home. We know a lot about different delivery systems, but not about the effect that searching has on the drop-off time, even though this is an essential element. There is no clear definition about searching itself, therefore, we will define it as “the process where the driver searches for a parcel inside

the truck.” In case that parcels are difficult to find, the drop-off time will increase and in

13

2.3 Scope

14

3 Methodology

The objective of this study is to explore the application of AR into last-mile logistics processes and measure its possible impact on efficiency. To this end, this study conducts a single case study and consists of two parts, namely verification and validation. The case study will provide insights into the complexity of applying AR to last-mile logistics (Karlsson, 2016). First, the application of AR was verified by creating a system design and was tested if it could perform the necessary functionalities in a virtual lab experiment. Second, a computational study was used to validate the effect when AR is applied to the efficiency of last-mile logistics. Simulation is the ideal method as it allows the researcher to adjust experimental factors (i.e. process times) and evaluate the outcome (Robinson, 2008).

3.1 Research setting

The case study was conducted at two companies as this would provide a thorough understanding from both the AR and last-mile logistics perspective. The first company is a leading innovator in the parcel delivery industry and is located in the Netherlands. They create best-in-class solutions in machine vision, recognition, autonomous sorting and identification technologies. A combination of their solutions allows an AR system to be successful. Multiple interviews and development sessions were conducted at this company with the R&D department and a computer engineer. During these sessions, information about the availability and maturity of AR was analyzed and discussed. Based on these sessions, a prototype of the AR-system was developed, and multiple tests were performed to verify its performance. The design of the AR-system was tailored to the needs of the parcel delivery company.

15 company was used to validate the possible effects that AR can have on the efficiency of the last-mile logistics.

The qualitative data collection had three goals. First, to get a thorough understanding of what an AR-system exists of and what it is capable of in last-mile logistics. Second, the data was used to create a realistic setting for the virtual lab experiment and to support findings during the experiments. Third, to assess whether the drivers would accept if someone else loads the truck.

3.2 Experimental setting

3.2.1 Virtual lab experiment

A prototype of the potential AR-system has been developed and a plan was created to test the two functionalities that the AR-system should perform. These functionalities were identified during the observations and interviews with drivers and experts in the parcel delivery industry. Tracking the location of the parcels and projecting this to the driver are two functionalities that have been tested. We tested the functionalities to measure the system’s readiness and verify if it can be adopted in the last-mile logistics. Multiple test runs were performed and a mean was taken from the results to provide a better estimate of the performance of the system (Robinson, 2008).

16 During the experiment, the functionality of the system was measured through observation and analyzed directly. In addition, screenshots were taken in order to prevent observer bias. The registration and tracking of the parcel were successful when a parcel got a unique code from the algorithm. This code had to match with the unique visual appearances of the parcels and an example can be found in Appendix B. If the parcel had the correct code, it was assigned with a 1 (successful) and when it was not the correct code with a 0 (unsuccessful). This allowed us to calculate the no. of times the system was successful in excel via =COUNTIF. The percentage of the system being successful was denoted as the success rate. Multiple experiments were conducted to test different types of algorithms used for tracking. This originated from continuously developing and testing the system to make it suitable for last-mile logistics. The verification of the AR system and its applicability in the last-mile logistics is presented and discussed in Section 4.

3.2.2 Simulation of the last-mile logistics

18

4 Findings

The following section presents the verification of the AR-system that is tailored for the last-mile logistics. The system is designed based on the requirements of the last-last-mile logistics are based on interviews, observations and literature review. First, the system design is presented and findings on how AR can be helpful in last-mile logistics. Second, the results of the experiments are presented and discussed.

4.1 AR-system design

Currently, the innovating company developed a prototype AR-system that can register and track the parcels inside the truck and projects the location to the driver via an AR-application. The AR-system consists of several components that can be divided into FRONTEND and BACKEND. The FRONTEND exists out of the physical elements (the mobile phone and the cameras) and the modeling systems (the AR-application and the XXX game engine). The mobile phone is used as a lens to project the information in an augmented form. The application is necessary to layer an augmented 3D arrow over the physical world. The AR-application is running on the XXX game engine in a c# programming environment. XXX is used to create 3D objects and this version can blend virtual objects seamlessly into the physical surroundings. The AR-application has a GPS tracker included that enables location tracking of the driver. The cameras are used for camera-vision to register and track the parcels inside the truck. The prototype uses one camera, the XXX, with 30 Frames-per-Second (FPS) to register and track the parcels inside the truck.

19

Figure 3 - AR-system design for the last-mile logistics

The developed AR-system is able to influence the loading process and the delivery process. During the loading process, the driver loads the truck using the LIFO method and optimizes the route. Drivers rely on their memory to find the location of the parcels inside the truck. This is an essential factor that ensures quick delivery. The information about the location of the parcels is crucial and the driver only knows this when he or she loads the truck. Therefore, loading and searching are interdependent. The AR-system can register and track the location of the parcels inside the truck through camera vision and store this information in the database. When the driver arrives at the customer, the AR-system gets a signal and knows that it arrived at the customer with a specific address and name. The tracking algorithm will then search for the designated parcel with the unique code and projects the location of the parcel via the AR-application to the driver. This creates the opportunity to decouple loading from the last-mile logistics.

20 last-mile logistics is shown in Table 1. This compares the current last-mile functionalities with the functionalities that an AR-system can perform.

Table 1 – applicability of the AR-system for last-mile logistics

Process Last-mile functionalities AR-system functionalities Loading The driver loads the truck via a LIFO method

and this will allow him/her to memorize where the parcels are located.

The location of the parcel can be registered through an object tracking algorithm (Karaca & Akinlar, 2005; Uma, 2019). Information can be stored and processed in a cloud (J. Lin et al., 2017).

Searching The driver uses his/her memory to remember where the parcel is located inside the truck.

The AR system can project the location of the parcel to the driver via the AR-application via an arrow (Lukosch et al., 2015; Mossel et al., 2012; Rejeb, 2019)

Delivery The driver has to scan the parcel for check-out.

The object recognition can be used to automatically scan parcel the parcels (Glockner, Jannek, Mahn, 2014; Rejeb, 2019; Uma, 2019).

The driver has to find the entrance to deliver the parcel(s).

Object recognition and historical GPS data can be used to help the driver find the entrance (Glockner, Jannek, Mahn, 2014).

The customer has to sign-off the delivery and verify his/her identity via their signature.

Facial recognition could be applied to identify customers and automate the sign-off the delivery (Glockner, Jannek, Mahn, 2014).

4.2 Tests

4.2.1 Tracking the location of the parcel

21 The state-based tracking algorithm uses states to track the location of the parcel. Each time a new parcel is loaded into the truck on a shelf, the current state changes and a new state is created. This state has a specific keyframe number (current keyframe number + 1) and is an image that is stored in a database. Figure 4 shows and example of the different keyframes during a run. The different states allow the algorithm to trace back the location of parcels that were loaded before because the system has a track record of all the different states. Thus, it is also aware of a parcel when it is not directly visible. In addition, state-based tracking is a common technique to reduce data redundancy and this reduces the computational power which is one of the technical challenges that prevent mainstream adoption. The tracking algorithm uses a search pattern and key points to find the parcel. Figure 5 shows the search pattern with the unique visual appearance and the key points that the algorithm uses to register and track the parcel. The prototype tests with a single keyframe had an average success rate of 80%, see Table 2 for the results of the experiments.

Figure 4 - key frames for single state-based tracking

22 state-based and the hybrid state-based. As visible, the hybrid state-based uses 74 keyframes to find the parcel and the single state-based has ten keyframes (See Figure 4).

Figure 5- search pattern (left side) and key points (right side)

Figure 6 - The view of the camera vision. Single state-based (left side) and hybrid state-based (right side).

23

Table 2 - results lab experiments

Run 1 Run 2 Run 3 Run 4 Run 5 Run 6 Average Single state (1 keyframe) Track 80% 70% 80% 90% 80% 80% 80% Project 70% 70% 80% 70% 80% 70% 73% Hybrid state (10 keyframes) Track 90% 100% 90% 100% 90% 100% 95% Project 90% 100% 90% 100% 80% 90% 92%

4.2.2 Projecting the location

The projection of the location of the parcel will enable the driver to find the parcel quickly. The user interface of the AR-application is shown in Figure 7. When the driver opens the application, he or she gets a popup with the delivery details and what the parcel looks like. After successful delivery, the driver can verify this and update the status (delivered or not delivered) into the delivery system. In addition, the location of the driver is calibrated in order to locate the parcel inside the truck accurately. The calibration happens with Visual-Inertial Odometry (VIO) and is the only viable alternative to GPS and lidar-based (Scaramuzza & Zhang, 2020). The AR-application uses the virtual marker “PrimeVision logo” to align the physical shelf with the virtual shelf. In addition, the driver will be able to finetune the calibration at the beginning of the shift in order to get perfect alignment. When calibrated correctly, the application scans the inside of the truck and presents the location of the parcel via an arrow, as seen in Figure 7.

25

5 Computational study

This section presents a computational study to validate the effect that AR can have on the efficiency of last-mile logistics processes. First, the computational design is explained, which is built on interviews, observations, literature and data provided by the logistics company. Second, the results from the computational study is presented to show where efficiency gains can be realized in the last-mile logistics processes when using AR.

5.1 Computational design

The objective of the computational study is to validate the effect of AR on the different processes in last-mile logistics. The last-mile logistics exist of fixed and variable processes and these have been translated into equation (1).

𝑡𝑡𝑜𝑡𝑎𝑙 = (𝑡𝐿𝑜𝑎𝑑1+𝑏× 𝑞 ) + (𝑡𝑟𝑜𝑎𝑑× 0,8𝑞 ) + (𝑡𝑠𝑡𝑜𝑝−𝑏 × 0,8𝑞 ) + 𝑡𝑆𝑒𝑡𝑢𝑝+ 𝑡𝐵𝑟𝑒𝑎𝑘

+ 𝑡𝑊𝑎𝑖𝑡𝑖𝑛𝑔+ 𝑡𝑈𝑛𝑙𝑜𝑎𝑑𝑖𝑛𝑔

(1)

Expression (1) represents the total time required by a driver to deliver q no. of parcels and is used as a measurement for the performance of the last-mile logistics. For simplicity, other indicators such as fuel costs or salary costs were not integrated. The first bracket represents the time for loading and is load-time per parcel times no. of parcels. The second bracket represents the transport between customers and is road-time times no. of stops. The third bracket represents the delivery and is stop-time times the no. of stops. In addition, the learning curve has been applied to 𝑡_{𝐿𝑜𝑎𝑑} and 𝑡_{𝑠𝑡𝑜𝑝} because this is affected by the driver’s experience. Where

b = learning curve = log Ø/log2, b < 0; learning rate, 0 < Ø < 1 (Kull et al., 2007). The learning

curve for 𝑡_{𝐿𝑜𝑎𝑑} is altered to 1+b because the load time is usually below 1 and now it shows the behavior accordingly. The learning curve for 𝑡𝑠𝑡𝑜𝑝 is normal because the stop-time is always

26 drivers is around 3,3 (Ø = 0,3) minutes and deviates 1,3 minutes from the average. The computational study will assign different values for these variable parameters because AR is expected to influence those.

For simplicity, fixed times were set for the way up and back to the delivery area and is denoted as 𝑡𝑠𝑒𝑡𝑢𝑝, waiting time denoted as 𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔, break denoted as 𝑡𝑏𝑟𝑒𝑎𝑘 or unloading denoted as

𝑡𝑢𝑛𝑙𝑜𝑎𝑑𝑖𝑛𝑔 between different routes. The reason behind this is only to test the effect that AR has

on the variable processes of the last-mile logistics. A driver’s workday is around 10 hours (600 minutes) and is used as a constraint in the model. This ensures that the increase in efficiency can only be achieved when time is saved in the current process. The vehicle capacity is affected by the sizes of the parcels and has an average maximum capacity of 240 parcels. The demand (d) for the depot center is 25.000 parcels a day. The no. of stops is denoted as 0,8q because sometimes multiple parcels have to be delivered at the same address. The average time per parcel is used to measure the no. of parcels that can be delivered extra or are lost. For simplicity, the average time per parcel is taken as the sum of the current average processing times 𝑡𝐿𝑜𝑎𝑑(0,8) + 𝑡𝑟𝑜𝑎𝑑(1) + 𝑡𝑠𝑡𝑜𝑝(2) = 3,8 minutes. The fixed parameters and base values represent

the current situation (base-case) at the warehouses and are determined by the interviews with the parcel delivery company.

Different values are used to illustrate the effect of AR on last-mile logistics performance. Two of the last-mile logistics processes have variable processing times. The simulation consists of appointing different values for these processes because AR could be used to decrease their processing times. The first variable processing time is loading and is denoted as 𝑡𝐿𝑜𝑎𝑑. The

second variable processing time is the delivery and is denoted as 𝑡𝑠𝑡𝑜𝑝. The observation and

interviews showed that AR could be used to decouple loading and this is simulated by a 𝑡𝐿𝑜𝑎𝑑

of 0. In addition, AR could reduce the delivery time by decreasing the time required for searching, scanning and finding the entrance.

27 their training sessions. Different values for experience (learning rate) were appointed to show this effect.

Eventually, there are two variable parameters that the current AR system is not able to exercise direct influence on. The first variable processing time is transport and is denoted as 𝑡𝑟𝑜𝑎𝑑. There

is less efficiency to be realized in rural areas because more time is spent on the road. The second variable parameter is the no. of parcels and is denoted as q. Different values for q shows the effect if managers want to increase the average truckload without changing time. In Table 4, an overview of the parameters, the notation, the base values and the experimental values is given.

Table 3- Input parameters for the experiment

Varying parameters Notation Base values Experimental values

Load-time in minutes 𝑡𝐿𝑜𝑎𝑑 0,8 0, 0.5, 1, 1.5

Road-time in minutes 𝑡𝑟𝑜𝑎𝑑 1 0.5, 1.5, 3

Stop-time in minutes 𝑡𝑠𝑡𝑜𝑝 2 1, 1.5, 2.5, 3

No. of parcels in units q 150 120, 150, 180, 210, 240

Learning rate Ø 0,60 .30 < Ø < .90

Fixed parameters

Demand d 25000

No. of stops in units x 80% of parcels

Transport depot to area and back 𝑡𝑠𝑒𝑡𝑢𝑝 60 minutes

Waiting time 𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔 20 minutes

Break 𝑡𝑏𝑟𝑒𝑎𝑘 30 minutes

Unloading 𝑡𝑢𝑛𝑙𝑜𝑎𝑑𝑖𝑛𝑔 20 minutes

Workday 592 minutes

Vehicle capacity 240

28

5.2 Computational Study

The computational study validates the effect that AR has on the efficiency of last-mile logistics. AR is considered to be able to reduce the processing times for loading and delivery because it is able to present information that can help them perform their tasks faster. In addition, AR is considered to be able to increase the experience of drivers and accelerate their training. These effects will be measured and compared with the performance of the current situation (base case). We measured the performance by three indicators and are explained below.

The first indicator is the reduction in total time that could be achieved through implementing AR in the last-mile and is measured by equation 2. The reduction in total time is an indicator of efficiency because the freed-up time can be used to deliver more extra parcels. This can lead to two types of efficiency improvements namely, a reduction in the no. of vehicles or by increasing the capacity of the current truck fleet.

𝑇𝑖𝑚𝑒 (𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜) − 𝑇𝑖𝑚𝑒(𝐵𝑎𝑠𝑒)

𝑇𝑖𝑚𝑒(𝐵𝑎𝑠𝑒) × 100%

(2)

The second indicator focuses on the reduction in the no. of trucks that could be achieved through implementing AR and is measured by equation 3. The reduction in the no. of trucks means that the current demand can be satisfied with a lower no. of trucks. Thus, resulting in lower salary and asset costs for the parcel delivery company.

𝑛𝑜. 𝑜𝑓 𝑡𝑟𝑢𝑐𝑘𝑠(𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜) − 𝑛𝑜. 𝑜𝑓 𝑡𝑟𝑢𝑐𝑘𝑠(𝐵𝑎𝑠𝑒)

𝑛𝑜. 𝑜𝑓 𝑡𝑟𝑢𝑐𝑘𝑠(𝐵𝑎𝑠𝑒) × 100%

(3)

The third indicator focuses on the capacity increase that could be achieved through implementing AR and is measured by equation 4. The capacity increase means that the current vehicle fleet capability to deliver extra parcels because AR was able to free-up time. This indicator could be used to prevent parcel delivery companies investing in new trucks to meet the growing demand but instead uses the increase in capacity.

𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦(𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜) − 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦(𝐵𝑎𝑠𝑒)

𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦(𝐵𝑎𝑠𝑒) × 100%

29

The effect on decreasing delivery time

Managers want to use AR to reduce the search-time, however searching is only a small portion of the activities during the delivery. However, this study shows that the effect of AR on the stop-time is limited and increases efficiency. In a scenario, where AR can assist the driver with finding the parcel and right entrance and automate the scanning and sign-off, it is likely to realize a reduction in average stop-time from 2 minutes to 1,5 minutes. As a result, the total time is reduced by 6% and can be used to deliver extra parcels. This offers parcel delivery companies with two choices. First, the parcel delivery companies can choose to reduce their no. of trucks with 6% resulting in lower operating costs. Second, the parcel delivery can choose to use the capacity increase of 7% to anticipate the growth in future demand. In this case, the parcel delivery company is able to deliver 1.570 parcels extra with the current fleet and thus no new trucks have to be acquired when the future demand grows. Therefore, using AR for the delivery phase can lead to efficiency improvements in last-mile logistics.

Table 4 - Stop-time scenario’s

Base Stop-time 1 1,5 2 2,5 3 Total time 512 554 592 628 662 No. of trucks 147 157 167 178 190 Capacity 171 160 150 141 132 Total time -14% -6% +6% +12% No. of trucks -12% -6% +7% +14% Capacity +14% +7% -5% -12%

The effect of increasing the driver’s experience

30 this time, the new drivers are not up to par and thus there is a loss in performance namely an efficiency gap. The analysis shows that a new inexperienced (learning rate 0,30) driver is only able to deliver 89 parcels within a 10-hour workday. An increase of 10% in experience (learning rate 0,40) will enable the driver to driver to deliver 119 parcels. A 10% increase in experience will reduce the efficiency gap with 20%. Therefore, it is crucial that the training of new drivers is completed as fast as possible to narrow the efficiency gap between new drivers and the base case. Figures 8 & 9, show the effect that a lower experienced driver has compared to the average time (learning rate <0,6). The graph shows that an inexperienced driver with a stop-time of 3 has the highest capacity loss. However, when the experience increases, the loss in efficiency decreases exponentially. Therefore, AR reality could be used to improve the efficiency of last-mile logistics as it speeds-up the training process and thus diminishes the effect of capacity loss as a result of the high staff turnover.

Table 5 - Learning rate scenarios

Base Learning rate 0,3 0,4 0,5 0,6 0,7 0,8 0,9 Total time 827 712 640 592 557 529 508 No. of trucks 281 211 182 184 158 151 146 Capacity 89 119 138 136 159 166 172 Total time +40% +20% +8% -6% -11% -14% No. of trucks +68% +26% +9% -5% -10% -13% Capacity -41% -21% -8% +6% +11% +15%

Figure 5 - Effect of inexperienced drivers on the capacity for delivery

-120% -100% -80% -60% -40% -20% 0% 20% 40% 0,3 0,4 0,5 0,6 0,7 0,8 0,9 Cap acity % Learning rate

Capacity loss inexperienced drivers (delivery)

31

Figure 6 - Effect of inexperienced drivers on the capacity for loading

The effect of decoupling loading

Implementing AR provides the opportunity to decouple loading and extend the tour length of drivers. To decouple loading someone else is required to load the truck and thus, extra employees have to be employed which will increase the operating costs at the warehouse. In a result, the driver spends their whole day delivering parcels and thus increases the utilization of the truck. Overall, the reduction in load-time increases the efficiency of last-mile logistics because more parcels can be delivered. What is surprising is the extent to which decoupling the loading process from the overall last-mile delivery has on efficiency. Decoupling loading, results in a reduction in a total time of 24% and a reduction no. of trucks by 26%. The reduction in total time is realized because the driver does not need to perform this activity anymore. The freed-up time leads to an increase of 25% in capacity and can be used to deliver 6.200 parcels extra with the current fleet. Therefore, using AR to decouple loading from last-mile logistics will have a significant impact on the efficiency of last-mile logistics.

Table 6 - Load-time scenarios

base Load-time 0 0,5 0,8 1 1,5 Total time 450 575 592 600 617 No. of trucks 134 163 167 169 174 Capacity 187 154 150 148 144 Total time -24% -3% +1% +4% No. of trucks -20% -2% +1% +4% Capacity +25% +3% -1% -4% -60,00% -50,00% -40,00% -30,00% -20,00% -10,00% 0,00% 10,00% 20,00% 30,00% 0,3 0,4 0,5 0,6 0,7 0,8 0,9 Cap acity % Learning rate

Capacity loss inexperienced drivers (loading)

32

6 Discussion and conclusion

This study explored the possibilities of applying AR in a last-mile logistics setting. It uncovered that AR could be an exciting technology to apply to a last-mile logistics setting. First, this study verified whether AR could be applied to last-mile logistics. After this we validated what effects AR can have on the efficiency of the last-mile logistics.

6.1 Theoretical implications

This thesis makes four theoretical contributions. This study contributes (1) to extend the discussion that AR can increase efficiency to the last-mile logistics, (2) it provides empirical evidence to support the debate on the effectiveness of AR, (3) it proposes a system design that would give a new definition to last-mile logistics, and (4) the role that AR can have on reducing the efficiency loss from staff-turnovers in last-mile logistics.

The first contribution of this thesis to literature is extending the debate of the effect AR on efficiency to the last-mile logistics. A number of scholars have shown that AR can increase efficiency in other logistics settings such as warehouses (Reif et al., 2010; Van Krevelen & Poelman, 2010). This study created an AR-system design specifically for last-mile logistics and showed that it is applicable in the last-mile context. The lab experiments showed that the system was able to perform the functionalities required in last-mile logistics. The system AR-system can track the locations of parcels, present the location to the driver and reduce the delivery time and thus increasing the efficiency of the last-mile logistics.

33

6.1.2 Last-mile logistics

The thesis also makes two contributions to the literature on increasing the efficiency of last-mile logistics by using a state-of-the-art solution. The first contribution of this paper is that AR can decouple loading from last-mile logistics and thereby freeing up time that could be used to deliver extra parcels. The lab experiment shows that the AR-system can decouple loading because it is able to register and track the parcels inside the truck by using camera-vision. The AR-system stores the information about the location of the parcels on a database and presents this to the driver when he or she opens the truck. The driver is guided to the designated parcel and is able to find the parcel without having loaded the truck. Thus, the driver’s time is used more efficiently, and it increases the utilization of the truck. As a result, it will increase the efficiency and redefine last-mile logistics as we know it because loading will become part of the warehouse activities. Last-mile logistics will then exist only out of transport and delivery which is a continuous re-occurring cycle. This outcome adds a new solution to the existing methods such as VRP, parcel lockers or collection points that are used to improve the efficiency of last-mile logistics (Iwan et al., 2016; Purnamasari & Santoso, 2018; Tsai et al., 2003). The findings of this paper can be used as a starting point to debate whether the last-mile logistics processes should be changed.

The second contribution to last-mile logistics is that the variable processes are affected by the driver’s experience. Experienced drivers perform tasks significantly faster compared to inexperienced drivers. The last-mile logistics is dealing with high staff-turnover and thus many inexperienced drivers need to be hired and trained. The variability in process times harms the efficiency because it makes accurate capacity planning more difficult. In addition, the new drivers need to be trained and this results in extra training costs. This study shows that AR can be used to diminish the negative effect of loss in capacity by accelerating the training of new drivers. Thus, it can reduce the variability of the process times and close the capacity gap rapidly. This paper adds another benefit to the discussion of why AR is a great tool to increase efficiency in logistics (Reif et al., 2010; Van Krevelen & Poelman, 2010).

6.2 Managerial implications

34 investments into new trucks. These insights can help parcel delivery companies to decide if they want to apply AR to their last-mile logistics processes. First, this study provides a system design for mile logistics. Specifically, a system that can decouple loading from the last-mile logistics. The simulation shows that the efficiency is mostly achieved when loading is decoupled from the last-mile logistics. However, companies could also consider other methods that are able to register and track the location of the parcels and present this to a driver in a different form. The latter could be achieved through a projecting system instead of an augmented reality application.

Furthermore, parcel delivery companies that deal with a high percentage of staff-turnover could benefit from AR because it can be used to reduce the variability between the driver’s performances. In addition, AR is able to accelerate the training of new drivers and thus to decrease the total training costs. The simulation shows that a significant efficiency gain can be achieved when the training of new drivers is accelerated. The reduction in variability will enable the parcel delivery companies to make more accurate capacity planning because there will be a lower gap between planned capacity and actual capacity. The abovementioned benefits could allow parcel delivery companies to either reduce their no. of trucks to meet the current demand or to use the capacity increase as a strategy to the constant growth in demand.

6.3 Limitations and Future research

35 Secondly, the effect of AR on the efficiency has been calculated via a simple simulation experiment where the average truckload was used in combination with average process times. This has been sufficient to roughly assess the possible impact AR can have on the last-mile logistics. In addition, this thesis used a simplified learning curve with small numbers and once it got closer to zero the effect would become linear. Future research should look into a more sophisticated model with other base numbers that are not close to 1. A suggestion would be to use seconds instead of minutes.

6.4 Conclusion

36

References

Allen, J., Thorne, G., & Browne, M. (2007). Good practice guide on urban freight transport. Bestufs

Administration Centre, 84. www.bestufs.net

Azuma, R. T. (1997). A survey of augmented reality. Presence: Teleoperators and Virtual

Environments, 6(4), 355–385. https://doi.org/10.1162/pres.1997.6.4.355

Bartholdi, J. J., & Gue, K. R. (2000). Reducing labor costs in an LTL crossdocking terminal.

Operations Research, 48(6), 823–832. https://doi.org/10.1287/opre.48.6.823.12397

Cai, Z., Chen, M., & Yang, L. (2011). Multi-source information fusion augmented reality benefited decision-making for unmanned aerial vehicles: A effective way for accurate operation.

Proceedings of the 2011 6th IEEE Conference on Industrial Electronics and Applications, ICIEA 2011, 174–178. https://doi.org/10.1109/ICIEA.2011.5975574

Cattaruzza, D., Absi, N., Feillet, D., & González-Feliu, J. (2017). Vehicle routing problems for city logistics. EURO Journal on Transportation and Logistics, 6(1), 51–79.

https://doi.org/10.1007/s13676-014-0074-0

Christopher, M. (2011). Logistics and supply chain management: creating value-adding networks. De Marco, A., Mangano, G., & Zenezini, G. (2018). Classification and benchmark of City Logistics

measures: an empirical analysis. International Journal of Logistics Research and Applications,

21(1), 1–19. https://doi.org/10.1080/13675567.2017.1353068

Demir, E., Bektaş, T., & Laporte, G. (2014). A review of recent research on green road freight transportation. European Journal of Operational Research, 237(3), 775–793.

https://doi.org/10.1016/j.ejor.2013.12.033

Ding, Y., Jin, M., Li, S., & Feng, D. (2020). Smart logistics based on the internet of things

technology: an overview. International Journal of Logistics Research and Applications, 0(0), 1– 23. https://doi.org/10.1080/13675567.2020.1757053

Fransoo, J. (2019). Curbside productivity for urban delivery needs to be improved: More parking space for delivery vehicles. Pulse, 1. https://www.linkedin.com/pulse/curbside-productivity-urban-delivery-needs-improved-more-jan-fransoo/

Gevaers, R., Van de Voorde, E., & Vanelslander, T. (2009). Characteristics of Innovations in Last Mile Logistics - Using Best Practices, Case Studies and Making the Link with Green and Sustainable Logistics. Association for European Transport and Contributors, October, 1–8. Glockner, Jannek, Mahn, T. (2014). Augmented reality in Logistics - Chaning the way we see

logistics (a DHL perspective). DHL Customer Solutions & Innovations, 20(3), 425–435. Hutchison, D. (2016). Augmented Reality, Virtual Reality, and Computer Graphics: 3rd International

Conference, Part I. In Augmented Reality, Virtual Reality, and Computer Graphics, Third

International Conference, Part I. https://doi.org/10.1007/978-3-319-40621-3_19

Iwan, S., Kijewska, K., & Lemke, J. (2016). Analysis of Parcel Lockers’ Efficiency as the Last Mile Delivery Solution - The Results of the Research in Poland. Transportation Research Procedia,

12(June 2015), 644–655. https://doi.org/10.1016/j.trpro.2016.02.018

37 with learning effects. International Journal of Production Economics, 131(1), 115–127.

https://doi.org/10.1016/j.ijpe.2009.04.019

Karaca, H. N., & Akinlar, C. (2005). A multi- vision system for real-time tracking of parcels moving on a conveyor belt. Lecture Notes in Computer Science (Including Subseries Lecture Notes in

Artificial Intelligence and Lecture Notes in Bioinformatics), 3733 LNCS(November), 708–717.

https://doi.org/10.1007/11569596_73

Karlsson, C. (2016). Research Methods for Operations Management. In Research Methods for

Operations Management. https://doi.org/10.4324/9781315671420

Kull, T. J., Boyer, K., & Calantone, R. (2007). Last-mile supply chain efficiency: An analysis of learning curves in online ordering. International Journal of Operations and Production

Management, 27(4), 409–434. https://doi.org/10.1108/01443570710736985

Lampropoulos, G., Keramopoulos, E., & Diamantaras, K. (2020). Enhancing the functionality of augmented reality using deep learning, semantic web and knowledge graphs: A review. Visual

Informatics, 4(1), 32–42. https://doi.org/10.1016/j.visinf.2020.01.001

Lin, C., Choy, K. L., Ho, G. T. S., Lam, H. Y., Pang, G. K. H., & Chin, K. S. (2014). A decision support system for optimizing dynamic courier routing operations. Expert Systems with

Applications, 41(15), 6917–6933. https://doi.org/10.1016/j.eswa.2014.04.036

Lin, J., Yu, W., Zhang, N., Yang, X., Zhang, H., & Zhao, W. (2017). A Survey on Internet of Things: Architecture, Enabling Technologies, Security and Privacy, and Applications. IEEE Internet of

Things Journal, 4(5), 1125–1142. https://doi.org/10.1109/JIOT.2017.2683200

Liu, F., Gui, M., Yi, C., & Lan, Y. (2019). A Fast Decomposition and Reconstruction Framework for the Pickup and Delivery Problem with Time Windows and LIFO Loading. IEEE Access, 7, 71813–71826. https://doi.org/10.1109/ACCESS.2019.2920444

Lukosch, S., Lukosch, H., Datcu, D., & Cidota, M. (2015). Providing Information on the Spot: Using Augmented Reality for Situational Awareness in the Security Domain. Computer Supported

Cooperative Work: CSCW: An International Journal, 24(6), 613–664.

https://doi.org/10.1007/s10606-015-9235-4

Milgram, P., Drascic, D., Grodski, J., & Restogi, A. (1995). Merging Real and Virtual Worlds.

Proceedings of IMAGINA, April 2018, 8.

http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:Merging+Real+and+Virtual+ Worlds#1

Mossel, A., Schönauer, C., Gerstweiler, G., & Kaufmann, H. (2012). ARTiFICe - Augmented Reality Framework for Distributed Collaboration. International Journal of Virtual Reality, 11(3), 1–7. https://doi.org/10.20870/ijvr.2012.11.3.2845

Moura, A., & Oliveira, J. F. (2009). An integrated approach to the vehicle routing and container

loading problems, Greece. In EURO XX-20th European conference on operational research. 1–

21.

Muth, E. J., & Spremann, K. (1983). Learning Effects in Economic Lot Sizing. Management Science,

29(2), 264–269. https://doi.org/10.1287/mnsc.29.2.264

Nathanail, E., Gogas, M., & Adamos, G. (2016). Smart Interconnections of Interurban and Urban Freight Transport towards Achieving Sustainable City Logistics. Transportation Research

38 Ong, S. K., Yuan, M. L., & Nee, A. Y. C. (2008). Augmented reality applications in manufacturing: A

survey. International Journal of Production Research, 46(10), 2707–2742. https://doi.org/10.1080/00207540601064773

Paelke, V. (2014). Augmented reality in the smart factory: Supporting workers in an industry 4.0. environment. 19th IEEE International Conference on Emerging Technologies and Factory

Automation, ETFA 2014, 1–4. https://doi.org/10.1109/ETFA.2014.7005252

Pierdicca, R., Frontoni, E., Pollini, R., Trani, M., & Verdini, L. (2017). The use of augmented reality glasses for the application in industry 4.0. Lecture Notes in Computer Science (Including

Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 10324 LNCS, 389–401. https://doi.org/10.1007/978-3-319-60922-5_30

Punakivi, M., Yrjo, H., & Holmstro, J. (2001). Solving the last mile issue : reception box or delivery

box ? 31(6), 427–439.

Purnamasari, C. D., & Santoso, A. (2018). Vehicle Routing Problem (VRP) for courier service: A review. MATEC Web of Conferences, 204, 1–9.

https://doi.org/10.1051/matecconf/201820407007

Ranieri, L., Digiesi, S., Silvestri, B., & Roccotelli, M. (2018). A review of last mile logistics innovations in an externalities cost reduction vision. Sustainability (Switzerland), 10(3). https://doi.org/10.3390/su10030782

Reif, R., Günthner, W. A., Schwerdtfeger, B., & Klinker, G. (2010). Evaluation of an augmented reality supported picking system under practical conditions. Computer Graphics Forum, 29(1), 2–12. https://doi.org/10.1111/j.1467-8659.2009.01538.x

Rejeb, A. (2019). The Challenges of Augmented Reality in Logistics : A Systematic Literature Review.

134(August), 281–311.

Robinson, S. (2008). Simulation: the practice of model development and use. Journal of Simulation,

2(1), 67–67. https://doi.org/10.1057/palgrave.jos.4250031

Ruano, S., Cuevas, C., Gallego, G., & García, N. (2017). Augmented reality tool for the situational awareness improvement of UAV operators. Sensors (Switzerland), 17(2), 1–16.

https://doi.org/10.3390/s17020297

Scaramuzza, D., & Zhang, Z. (2020). Aerial Robots, Visual-Inertial Odometry of. Encyclopedia of

Robotics, 1–9. https://doi.org/10.1007/978-3-642-41610-1_71-1

Schwerdtfeger, B., & Klinker, G. (2008). Supporting order picking with augmented reality.

Proceedings - 7th IEEE International Symposium on Mixed and Augmented Reality 2008, ISMAR 2008, 91–94. https://doi.org/10.1109/ISMAR.2008.4637331

Sisto, M., Wenk, N., Ouerhani, N., & Gobron, S. (2017). A study of transitional virtual environments.

Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 10324 LNCS(June), 35–49.

https://doi.org/10.1007/978-3-319-60922-5_3

Stoltz, M. H., Giannikas, V., McFarlane, D., Strachan, J., Um, J., & Srinivasan, R. (2017).

Augmented Reality in Warehouse Operations: Opportunities and Barriers. IFAC-PapersOnLine,

50(1), 12979–12984. https://doi.org/10.1016/j.ifacol.2017.08.1807

39

traveling salesman problem. 20(4), 179–186.

Uma, S. (2019). Latest Research Trends and Challenges of Computational Intelligence Using Artificial Intelligence and Augmented Reality. Computational Intelligence and Sustainable

Systems, 43–59.

Uygun, Y., & Reynolds, E. B. (2017). Advanced Manufacturing Innovation Ecosystems: The Case of

Massachusetts. https://doi.org/10.1007/978-3-319-42559-7_29

Van Krevelen, D. W. F., & Poelman, R. (2010). A Survey of Augmented Reality Technologies, Applications and Limitations. International Journal of Virtual Reality, 9(2), 1–20.

https://doi.org/10.20870/ijvr.2010.9.2.2767

Veenstra, M., Roodbergen, K. J., Vis, I. F. A., & Coelho, L. C. (2017). The pickup and delivery traveling salesman problem with handling costs. European Journal of Operational Research,

257(1), 118–132. https://doi.org/10.1016/j.ejor.2016.07.009

Wang, Y., Zhang, D., Liu, Q., Shen, F., & Lee, L. H. (2016). Towards enhancing the last-mile delivery: An effective crowd-tasking model with scalable solutions. Transportation Research

Part E: Logistics and Transportation Review, 93, 279–293.

https://doi.org/10.1016/j.tre.2016.06.002

Zobel, B., Berkemeier, L., Werning, S., Vogel, J., Ickerott, I., & Thomas, O. (2018). Towards a modular reference architecture for smart glasses-based systems in the logistics domain. CEUR

40

Appendices

Appendix A

41 Appendix B

Virtual Lab Experiment 7 July 2020

Participants:

A: The computer engineer and driver B: The researcher

Objective:

The objective of this experiment is to measure two functionalities that the AR-system has to replace in order to be an ideal solution for last-mile logistics. These two functionalities are:

1. Registering and tracking the location of the parcel inside a truck 2. Projecting the location to the driver via a mobile phone application

The reason why we want to test these two functionalities is that within last-mile logistics, the loading phase (performed by the driver) is tightly coupled to the search phase (performed by the driver). The driver loads his/her truck and registers where the parcel is located inside the truck. When arriving at the customer the driver is using this information to find the parcel.

Requirements (materials, setup etc):

The requirements to make the virtual lab experiment possible are: 1. Internet connection via teams

2. Two testers (A in Romania and B in the Netherlands) Materials:

1. 10 (each different addresses, attributes and appearance) (no more address examples) a. Different addresses & boxes

2. Shelf setup (replicating the inside of a truck) 3. Extra lightning (to ensure registering)

4. Stand for pre-loading and delivery (where the boxes are placed before loading) Hardware: 1. Mobile phone 2. Modem 3. 2 laptops 4. Camera Software: 1. XXX for database

2. XXX game engine for projecting augmented reality

3. Dashboard (web) app to replicate the route called the Taskboard (see below for illustration) 4. Algorithms for

a. Tracking b. Projecting

42

Experiments:

To test if the system is able to perform these functionalities we will perform tasks that replicate each step of last-mile logistics. There are two steps that are investigated namely: loading and searching. Each step and the activities are explained below:

Loading:

1. Pick up the first parcel

2. Scan it (put it on the shelf. Give the sensors time to register it)

3. Pick up the 2nd _{parcel and perform the same task until the last parcel is loaded}

4. Close the door 5. End

Searching:

1. Open the AR application

2. Accept the new address with a designated parcel 3. Point the camera to the shelf

4. Pick up the parcel that has an arrow pointed at it 5. Put the parcel on the delivery stand

6. End

Data recording:

To test whether the functionalities are successfully performed we will record the data in the following way:

1. Via screenshots we will be able to record the tracking and registering algorithms in the truck 2. Via screenshots we will be able to record the projection of the arrow

3. Via Excel, we will keep a log book of the multiple runs (6 runs) we would perform a. We will test whether the system registers each parcel

b. We will test whether the correct parcel is picked up

c. We will create data tables and graphs to illustrate the success rate of the system i. The no. of attempts the system functions accordingly

Analysis

The analysis will be conducted on the data that is collected during the experiment. The analysis will be describing:

1. Whether the system was able to register the parcels and the success rate of this (i.e. 80 out of 100 meaning a success rate of 80%)

2. Whether the system was able to project the location of the parcel and the success rate of this (i.e. 80 out of 100 meaning a success rate of 80%)

3. Explain how the experiment went, what went well, complications that occurred, what the limitations are, and the outcome of the test

Example photo pictures addresses

Task board:

44 Appendix C

Results from virtual lab experiments

46