Section 2

Methodology

The how — approach, sample, project timeline, and the four research methods used.

Section 2.1

Approach

Qualitative and quantitative, weighted toward qualitative because the unit of analysis — the handoff moment — is behavioral and emotional, not numerical.

Why ethnography, not surveys

A survey would have told us how many people say they want robotic delivery. The handoff is a moment that lives in behavior, not self-report.

What the methods observe

  • Body language at pickupVisible only in shadowing.
  • Confusion with the unlock gestureVisible only in cultural probes.
  • Sensory disorientation on a sidewalk in front of strangersVisible only in intercepts.
  • Restaurant workflow micro-decisionsVisible only in restaurant interviews.

Why both qualitative and quantitative

Qualitative work surfaces the shape of the problem (where, with whom, why). Quantitative work supplies the magnitude (how often, how costly, how many). Every qualitative finding is grounded in a numeric data point where one exists, and every numeric data point is illustrated by a qualitative quote.

The impact–effort matrix

How findings get prioritized

ImpactHow much it would move the first-time-use (FTU) adoption needle, weighted by the operational cost of the failure.
EffortEngineering, design, partnership, and policy lift.

The matrix prioritizes the upper-left quadrant (high impact, low effort) and demotes the lower-right. The final recommendations in § 5.2 are filtered through this matrix.

Methods used

22 primary touchpoints across 13 matrix rows

MethodTypenPurpose
Online researchSecondary—Market context, sustainability data, infrastructure data
Individual interviewsPrimary · qualitative11The four audiences (see § 2.2)
Cultural probesPrimary · qualitative + photo/video5 participantsReal-world ordering attempts with documentation
Sensory cues interceptsPrimary · qualitative6 interceptsImmediate emotional + sensory signals near a robot
ShadowingPrimary · observationalseveralRobot behavior + bystander reaction in deployment zones

What the approach is not

  • Not representative.The sample is purposeful, not random.
  • Not statistically significant.Findings are directional evidence for design decisions, not population-level claims.
  • Not a usability test of any specific app.App findings emerged organically from the cultural probes; they were not the primary target.

Ethics

  • Consent collected verbally at the start of every interview and intercept.
  • Cultural-probe participants signed a written consent for photo/video.
  • Names are used in the matrix only with explicit permission. Where participants wished to remain anonymous, they are referenced by role.

Section 2.2

Sample & recruitment

10–12 one-on-one qualitative interviews across four audience groups, each surfacing a different angle of the same handoff moment.

Overview

The team conducted 10–12 one-on-one qualitative interviews with individuals matching our audience criteria, split between four groups. Findings inform a research project investigating the robot-to-human handoff moment in autonomous sidewalk food delivery.

Research goals

  1. 1.Understand the emotional and functional elements of the hand-off moments with robotic food delivery.
  2. 2.Understand the interaction between robot and customer, customer perception, and customer experience with robotic food delivery, apps, and integrated systems.
  3. 3.Identify the outcomes that restaurants get with robotic food deliveries.
  4. 4.Identify the friction points between robots, customers, apps, and restaurants.
  5. 5.Surface changes in workflow or routine caused by robotic delivery.
  6. 6.Understand expectations of trust and food quality.

Audience segmentation

  • Customers with experienceClients who have received at least one robotic food delivery.
  • Customers without experienceClients familiar with food delivery apps but no robotic-delivery experience.
  • Restaurants with robotic serviceRestaurants currently dispatching to delivery robots (e.g., Uber Eats / Serve Robotics partners).
  • Restaurants without robotic serviceRestaurants using only human couriers.

Each group surfaces a different angle of the same handoff moment: dispatch friction (restaurants with), expectations and gaps (restaurants without), real-world handoff (customers with), and adoption barriers (customers without).

Section 2.3

Project timeline

Eight weeks (W3–W10) across six phases. Deliberate parallelism between fieldwork and analysis.

8Weeks activeW3–W10
6Phases
2Class meetings / weekMon + Thu
May 28DeliverySpring 2026

The project ran across eight weeks of the ten-week quarter, structured in six phases with deliberate parallelism between fieldwork and analysis. Phases are tight on the front end (definition) and at the back end (synthesis); the middle weeks accept overlap because that’s where research data and analysis cross-pollinate in practice.

Phase view

Six phases across W3–W10

PhaseWeeksDatesFocusMilestone
1 · DiscoveryW3Apr 6 – Apr 12Topic explorationTopic locked
2 · DefinitionW4Apr 13 – Apr 19Research statement · Recruiting kickoffRS approved · Recruit list opened
3 · Fieldwork (parallel)W5–7Apr 20 – May 1011 interviews · 5 cultural probes · 6 sensory cues · online research · shadowingAll raw data collected
4 · AnalysisW7–8May 4 – May 17Affinity diagram · Data filtering · Lextant populationIdeal Experience drafted
5 · SynthesisW8–9May 11 – May 24Final conclusions · Highlight prioritization · Client book templateOutline locked
6 · DeliveryW10May 25 – May 28Final client book build & handoffClient book shipped (Thu May 28)

Class-day view

Two class meetings per week, anchoring milestones

W3
Mon Apr 6Thu Apr 9
Topic exploration kickoff; team forming
W4
Mon Apr 13Thu Apr 16
Research statement reviewed; recruiting plan approved
W5
Mon Apr 20Thu Apr 23
Interview kickoff (Alejandro begins customer-side intercepts)
W6
Mon Apr 27Thu Apr 30
Mid-project check-in; cultural probes deployed
W7
Mon May 4Thu May 7
Sensory cues intercepts at GT campus; data collection wraps
W8
Mon May 11Thu May 14
Affinity wall built; Lextant framework populated
W9
Mon May 18Thu May 21
Outline locked; client-book template approved
W10
Mon May 25Thu May 28
Final client book delivered

Why this sequence

Why front-load definition (W3–W4)A precise research statement is worth more than a rushed start. Locking Intent · Impact · Method before fieldwork prevents drift mid-project.
Why parallel fieldwork + early analysis (W6–W8)Qualitative research yields signal earlier than people assume. Starting affinity work in W7, while sensory cues were still being collected, let us adjust later interviews to probe gaps.
Why a dedicated synthesis phase (W8–W9)Without explicit time for writing and prioritization, fieldwork artifacts never become a client deliverable.
Why a single delivery week (W10)The client book ships once. The last week is dedicated to assembly, review, and final QA — not new research.

Fieldwork detail (Phase 3)

What happened during W5–W7

ActivityWindowLeadNotes
11 interviewsW5–6Vicente, Jacob, Alejandro4 audiences (see § 2.2)
5 cultural probesW6–7TeamCamille, Nella, Heidi, Hanalai, Taner — each with notes + photos + video
6 sensory cues interceptsW7TeamRandom pedestrians, Georgia Tech campus
Online deep researchW6–7Team (parallel)Market data, regulation, sustainability, infrastructure
ShadowingW6–7TeamObservation of robot deployments at GT and Beltline

Risks managed during the timeline

Probes returning thin dataMitigated by the sensory-cues fallback exercise (which became the W7 work in its own right).
Recruiting lagStarted recruiting in W4, ahead of W5 interviews.
Analysis-time crunchStarted analysis in parallel during W7, not after fieldwork closed.

Section 2.4

Primary research

The original data the team collected — four methods, 22 touchpoints across 13 matrix rows. This is where the central argument of the book lives.

The original data the team collected during the project window (Apr 6 – May 28, 2026). Four methods, 22 touchpoints across 13 research matrix rows.

Method 01

Individual interviews

n=11 · Semi-structured · 30–45 min · Audio-recorded with consent.

  • Two formats in parallel — pre-scripted structured interviews and spontaneous intercept interviews near deployment zones (Atlanta Beltline, Georgia Tech, Virginia Highland).
  • 7-step interview frame — opening + consent → warm-up → robot-specific probe → handoff moment → trust + food quality → workflow / repeat-use → closing.

Result

Audio recording → highlights extraction → attributed insights and quotes added to the research matrix and the affinity wall.

Method 02

Cultural probes

5 participants · Real-world ordering task with notes, photos, video · Structured debrief.

  • Participants: Camille Armstrong, Nella Kuse, Heidi Hampton, Hanalai Hampton, Taner.
  • Documentation kit: brief, consent form, prompts, post-activity interview structure.

Result

The single strongest finding of the project: 5 of 5 motivated participants failed to complete the task. None could surface a robot option in mainstream apps. The handoff moment never happened — the failure was upstream.

Method 03

Sensory cues intercepts

n=6 · Brief intercepts on Georgia Tech campus paired with observation of robot interactions.

  • Cultural probes surfaced upstream failures (discoverability) but couldn’t capture sensory signals around a visible, present robot. Sensory cues filled that gap.

Result

Dominant emotional register toward sidewalk robots is neutrality. Brand identity does not survive the trip from app to sidewalk. By the third encounter, the robot is invisible.

Method 04

Shadowing & field observation

Notes only · No participant intervention.

  • Observation of robot deployments at Georgia Tech and the Atlanta Beltline — robots in motion, pedestrians negotiating shared space, restaurant staff loading robots.

Result

Notes feed the Last meter constraints and Restaurant loading sections of the analysis; specifically informed the infrastructure & terrain gaps line of the secondary-research synthesis.

Synthesis flow

Raw data → recommendations

  1. Raw dataaudio, video, photos, notes
  2. Highlight extractionquotes pulled with attribution
  3. Research matrix13 rows · structured insights + quotes
  4. Affinity diagram7 columns · 3 layers (ideal · current · raw)
  5. Lextant populationIdeal Experience framework
  6. Personas + OpportunitiesRecommendations

Section 2.5

Secondary research

Existing data the team synthesized to set the macro context behind primary findings — market, cost, throughput, sustainability, infrastructure, safety, behavior, intervention triggers.

Demand and market

How big the food-delivery market is and how fast it's growing.

+63%Pandemic delivery uptick
185M+U.S. users projected by 2025
12×Uber Eats revenue growth ($1.1B → $12.1B, 2017–2023)
~24%Uber Eats U.S. food delivery share
$8.9BFood robotics market by 2033
150%YoY robotic delivery growth (2024)

Cost structure

Where each delivery dollar goes and what robots change.

~80%Driver labor share of delivery cost
~40%Cost-per-order reduction with robots
20–25%Labor cost drop (largely offset by teleoperator salaries)
GapIndustry goal: 1 teleoperator : 20 robots. Urban reality: 1 : 5 due to cognitive load.

Throughput

How many orders a single human or robot moves per trip.

3–5Orders per human-driver trip
1Order per robot trip
<10%Robot active-duty cycle today
1.5–3 miTypical delivery distance

Sustainability

Per-delivery emissions vs. gas vehicles and system-level impact.

67–99.9%Per-delivery emission reduction vs gas vehicles
612.7gCO₂/mi gas vehicle vs 9.3g robot
16%System-level CO₂ reduction
29%Urban congestion reduction
#1Lowest-carbon mode for grocery + food delivery

Infrastructure

How well today's streets and buildings support delivery robots.

30–40%Path unmappable in deployment zones
25%Performance drop with non-standard obstacles
#1IoT-elevator integration: scaling requirement 2026
~90%Existing doorbells / handles robots can't engage
15sSignal loss that can trigger manual rescue
~58%Outdoor sidewalk robot market share
GapHybrid all-terrain units growing fastest at 27.8% CAGR.

Safety

How robots interact with pedestrians on shared sidewalks.

40Dangerous near-misses (NAU 5-day study)
60Moderate-risk pedestrian–robot interactions
GapRobots were to blame in most cases.

User behavior (campus benchmark)

How students using campus delivery robots behave once habituated.

61%College students placing weekly robotic ordersStarship national campus survey 2024, n = 7,063
98%Positive sentiment among active users
66%Users skipping fewer meals due to access

Intervention triggers

What forces remote operators to take over from autonomous mode.

70%Of human interventions caused by connectivity + blocked routes
GapStructural and signal infrastructure, not AI software.

Up next

Section 3 · Analysis Framework

From raw transcripts to structured insight — coding, affinity diagramming, and the Lextant translation to drivers.