From bicycles to robots: acceptance of automated technologies for parcel delivery

From bicycles to robots: acceptance of automated technologies for parcel delivery
Like

An Unexpected Journey from Bicycles to Autonomous Delivery

What do bicycles have to do with research on automated delivery technologies? The conception of this research holds its origins in what was supposed to be research on bicycle parcel delivery in Germany.

The original idea started from a collaborator reaching out with data on bicycle parcel delivery, seeking analytical expertise. As we geared up to work on the data, with funds already allocated to the project, we learned of Ford’s work on a delivery robot, Digit. Having already established the momentum to work on the topic of parcel delivery, that fusion of automation and parcel delivery proved too intriguing to pass up: the idea for this work was born and became the cornerstone of my Ph.D. dissertation.

Consider these staggering figures: In 2022, over 161 billion parcels were shipped globally. By 2027, this number is projected to soar to 256 billion[1]. Meanwhile, 2022 saw retail e-commerce sales surpass 5.7 trillion U.S. dollars[2]. The critical, yet inefficient, final leg of these deliveries – known as last-mile logistics – represents up to 28% of transportation costs[3], contributing significantly to urban congestion, pollution, and noise.

This research with my co-authors, Amanda Stathopoulos and Spencer Aeschliman, pivots around a key but often overlooked stakeholder in this process: the customer. We explore not just the acceptance of four automated parcel delivery technologies but also the collective perception of these innovations under the autonomy umbrella.

Automated delivery technologies assessed in this study – autonomous vehicles, drones, sidewalk robots and bipedal robots

Pushing the Boundaries of Modeling and Computing

Focusing on the analysis of four emerging delivery technologies – (1) autonomous vehicles, (2) drones, (3) sidewalk robots and (4) bipedal robots – our study presented an ideal opportunity to stretch our modeling capabilities. We developed a comprehensive model encompassing choice, nesting, and correlated latent variable components. On a standard computer, this model would require 10 days for estimation. However, by leveraging smart computational strategies and Northwestern University's high-performance computing resources, we dramatically reduced this timeframe.

Key to our approach was customizing `apollo`, an R package for choice modeling, to streamline setup, output retrieval, and results summarization. We opted for Sobol draws over traditional Halton draws for better performance in higher dimensions and employed antithetic draws, which expedite model convergence. These improvements, combined with the power of high-performance computing (utilizing up to 128 CPU cores), significantly accelerated our process.

This computational efficiency enabled rapid testing and development of new models. Models that once took days to run were completed in hours, and we could run multiple models simultaneously. A unique tool we developed tracked parameter stability across models, boosting our confidence in the results' accuracy and convergence. Ultimately, we could afford a higher number of draws for estimation, enhancing model precision.

After running thousands of model permutations over the span of a year, the final version, using 20,000 draws, converged in just 18 hours on 28 CPU cores.

Sample from choice model stability testing (20 runs with randomly selected starting values using 500 draws)

Insights into Customer Preferences and Acceptance

Our findings reveal nuanced preferences among emerging delivery technologies. Cost and time efficiency are key factors driving customers toward automated options, indicating a readiness to embrace new delivery methods under optimal delivery conditions. Interestingly, drones stood out as a particularly favored option in scenarios that balanced performance with cost.

We observed a differentiation between autonomous vehicles and smaller-scale technologies like sidewalk and bipedal robots, with the latter being perceived as more interchangeable. The study also highlighted the value of time in accepting automated deliveries, underscoring the importance of convenience in these rollouts.

Presence requirements during delivery varied across technologies. While traditional methods showed little sensitivity to recipient presence, autonomous vehicles and drone deliveries displayed a marked preference for the recipient's presence, reflecting security concerns.

The study also considered the impact of parcel content on acceptance. High-value items, like smartphones, garnered less trust in automated delivery, whereas groceries were more positively associated with self-driving cars, reflecting concerns over handling perishable goods.

Socio-demographic factors, including age and gender, also influenced technology acceptance. Older individuals and women exhibited less inclination toward automated options, while higher acceptance was noted among those with graduate-level education and Asian and Pacific Islander respondents.

Toward a Future of Automated Deliveries

Our research highlights the intertwined roles of attitudes toward technology and package handling concerns. A higher affinity for technology correlates with a preference for automation, while package handling concerns temper acceptance of automated modes. These insights indicate that addressing parcel-handling trust issues could be key to catalyzing wider acceptance.

Scenario simulations shed light on potential market responses to various delivery attributes and policy interventions. Strategies like competitive pricing, improved service, and promotional efforts emerged as effective levers for encouraging the adoption of automated delivery technologies.

In sum, our research journey, from bicycles to robots, opens a window into the complex, evolving landscape of automated delivery. It's a journey that reflects the dynamic interplay between technology, consumer preferences, and the future of logistics.

Finally, we leave with you a few amusing highlights from the open-ended survey responses. On bipedal robots, one respondent feared for their children and pets, “The bi-pedal delivery is a bit creepy and would probably scare my child and dogs.” Some did not jive with these technologies at all, “I do not want any of those freaky deaky things bringing packages to my home. I want a person. Period.” Yet, warming our hearts, some found appreciation in these technologies, “The sidewalk robot is so cute!!! It's adorable!!! I would trust this robot with my LIFE.”

Word cloud for open-ended survey responses on opinions towards automated delivery technologies (green = positive sentiment, red = negative sentiment)
Word cloud from open-ended survey responses on opinions towards automated delivery technologies
(green = positive sentiment; red = negative sentiment; asterisk indicates negation)

[1] https://www.statista.com/statistics/1139910/parcel-shipping-volume-worldwide/

[2] https://www.statista.com/topics/871/online-shopping

[3] Ranieri, L., Digiesi, S., Silvestri, B. & Roccotelli, M. A review of last mile logistics innovations in an externalities cost reduction vision. Sustainability 10, 782 (2018).

 

Please sign in or register for FREE

If you are a registered user on Civil Engineering Community, please sign in

Related Collections

With collections, you can get published faster and increase your visibility.

Retinal imaging and diagnostics

This Collection invites works providing insight into using novel or existing retinal imaging technologies in clinical applications or presents new or adaptive forms of these techniques.

Publishing Model: Open Access

Deadline: Jan 31, 2024

Pharmacogenomics for precision medicine

This Collection presents recently identified genomic pathways, novel genetic biomarkers, and epigenetic signals influencing treatment outcomes for particular patient populations.

Publishing Model: Open Access

Deadline: Ongoing