LaundryLoopYiting WangCOMP1110 · GROUP 24

LaundryLoop.

A Project about User-Reported Coordination
in Shared Laundry Facilities

Group 24
Shuyue Cheng · Chenyang Hu · Yiting Wang
Chenxi Zhang · Rongyu Zhang

PEOPLE
IN THE LOOP
Yiting Wang
BASED ON HIGH WEST RESIDENT EXPERIENCES

Starting from frustration in the laundry room:
Why can’t we do laundry smoothly?

Resident A 😣Guys, washer 1 & 4 have finished already.
Please collect your clothes on time!
Resident B 🥀Tired of waiting for people to pick up their clothes…
Resident C 🤔Is the machine’s countdown really accurate?
Resident D 💔I was only a few minutes late. Someone took my clothes out and left them in the dusty collection basket.
One machine out of service
A room-wide disruptionPayment-system failure reported by residents

Translated from Chinese. Anonymised for privacy.

Yiting Wang

Coordination is hard because:

Delayed collection
machine unavailable
transfer from washing to drying
delayed

Moving clothes is a direct but
imperfect solution

Machine status isn’t reliably shown
online

Yiting Wang

OUR QUESTION:How far can credit-based allocation, reminders and collection coordination reduce queueing when machine status depends on resident reports?

Yiting Wang

Existing research
approaches

The gap

Sensor Systems

Work by measuring machine states

Jo et al., 2019; Portase et al., 2024

Booking Platforms

Work with slots & notifications

Marcu & Damian, 2025; Serajudin & Mostafa, 2025

Behavioural designs

Work on incentives and shared expectations

Lam et al., 2016; Fossett, 2026
THE GAP

Coordination of the system


OUR SETTING

A system based on
incomplete &
self-reported information

Yiting Wang

Design

The Three-step Loop

01

Report

Scan QR code to start and collect

02

Coordinate

System allocates the queue & reminds you.

03

Correct

Found clothes in a “free” machine?
Report it! The machine turns “Unknown”.

Chenxi Zhang
FROM RESERVATION TO COLLECTION

One complete laundry journey.

OPTIONALReserve a washer
ALLOCATIONCredit + waiting
Or use an available washer directly ↙
01 · SCAN TO STARTWash — 35 minRegister machine use
02 · SCAN TO COLLECTCollect & transferRelease the washer
03 · REGISTER USEDry — 5–60 minSubject to availability
04 · SCAN TO COLLECTFinishRelease the dryer
Optional linked dryer request
Report a mismatch → Check & correct availability
Chenxi Zhang
DESIGNED FOR PEOPLE

Beyond the Algorithm

Algorithms can’t solve everything.

Friendly design encourages considerate behaviour.

Chenxi Zhang
THE FIRST DRAFT

A quick flashback

A simple simulation of the LaundryLoop proposed in our first draft.

There are some major differences compared with the current design:

  1. Users choose a time range, rank the machines and apply a day before their laundry day.
  2. Gale–Shapley stable matching matches students to machines in the simulated time period. Each machine matches five students.
  3. The system determines the sequence and time in which students can wash their clothes.
  4. The reminder algorithm introduced in “Future” decides when to send “You may come down now”, aiming to maximise efficiency and make the machine very likely to be empty when the next user arrives.
Chenyang Hu
EARLY ALGORITHM EXPLORATION

Stable matching

Rank machines → Propose → Keep or reject → Match

Chenyang Hu
BEFORE CONFIRMING A HOLD

Which machines are available?

01

Fresh vacancy report

Fresh means ≤ 10 min old

02

No active session

In the applying user’s schedule

03

No conflicting confirmed hold

We can’t arrange the machine for two people

04

Machine not damaged

Reported by users

05

No uncertain state

Or disputed — almost the same here

The system predicts cycle finishing times by adding 35 minutes to the user-reported start time.
A finished cycle does not confirm an empty machine: clothes may still be waiting for collection.
User-sent collection signals help confirm vacancy. However, there are tricky users…

Chenyang Hu
CREDIT + WAITING COMPENSATION

Queueing rule

Pi=Ci+Wi5
User 1: 96 credits + 20 minutes waited96 + 4 = 100Considered first
User 2: 98 credits + 0 minutes waited98 + 0 = 98Sorry, have to wait
>

2 additional rules:

  1. Confirmed holds stay in place.
  2. “Waiting time” starts from limited eligible waiting time.
Applying 22 hours earlier won’t make you
have some sort of undefeatable priority
Chenyang Hu
VISIBLE RULES

Credit feedback rules

Unlike Alipay, credits and rules are transparent to users.We’re quite proud of this.

On-time scanned collection+2
Collection over 10 minutes late−3
Manual collection−1
Reservation no-show−8

Credits start at +100 and stay between 0–120.
Positive rewards are limited to +4 per day.

Chenyang Hu

Two Stages

To connect the dryer:

35 minWashing
→
2 minTransfer
→
5–60 minDrying

To kill processes:

15 minHow long a confirmed hold stays valid120 minAfter this, an unsuccessful request is cleared from the system
Chenyang Hu

Future: Two extensions to help with peaks

Wait, how can I explain all these in 30s? 💡 I can just put them here.

1. Earlier reminders

Normal-distribution reminder algorithm with shaded arrival and collection probabilities
C ∼ N(μC, σC2)
A ∼ N(μA, σA2)
Conflict event: A ≤ t < C
P(conflict at t) = P(A ≤ t) · P(C > t)
P(conflict at t) ≤ 0.05
Tcheck = max(μC + 4σC, 36 + μA + 4σA)

This sets an upper limit to the time we search for t.
If no t is found, use this upper limit.

2. Personal recommendations

S=LDemand+DDue time+HHabit+JSmall offset
S
Total score for a recommended time.
Lower = better
L
Estimated pressure on laundry resources
D
Absolute distance from the estimated due time, to the end of the user’s chosen time period
H
Mismatch with the resident’s usual clothes-washing time
J
Deterministic resident-specific offsets, with magnitude ≤ 0.1, to avoid recommending the same time to everyone
Chenyang Hu

Experimental Results

Mean queueing time per completed wash-and-dry load in a simulated laundry room.

On-site waitingRemote waitingMean minutes / load
0255075100 min

420 residents · 10 washers + 10 dryers · 30 paired seeds · Percentages denote low-reliability residents.
95% confidence intervals · Synthetic results · Cycle durations excluded.

Rongyu Zhang
FROM SIMULATION TO A RESIDENTIAL PILOT

The next step

What remains open

Synthetic behaviour

Reporting and collection vary together

One-room setting

One room in one semester.

Measure report freshness and collection

Test features separately

Evaluate the proposed extensions

Promising simulation results → Refine through a real pilot

Shuyue Cheng

YOUR TURN toExplore the prototype

Demonstration data.
No registration required.

Open the live demo ↗
Shuyue Cheng
Original photograph supplied by the team