For the first time ever, Roombas learned to predict cleaning needs and act automatically, eliminating floorcare maintenance from users' daily routines.
Freeing Roomba Users From Floorcare with Automated Cleaning
PROJECT OVERVIEW
Building an experience around cleanliness predictions unlocked the ability for Roombas to automatically decide which rooms needed to be cleaned and how best to clean them. This introduced the ability for users to set it and forget it — leaving their Roombas to automatically decide if rooms needed to be cleaned, and prioritize them based on cleanliness.
This feature is rooted deeply in user feedback, and was a huge step towards a truly autonomous, goal-driven robot rather than an appliance that waits to be told what to do. It shipped Q1 2024 as iRobot's first intelligent, predictive feature and marked a significant departure from the robot-centric app experience.
I led the design effort on a team of product management, research, and engineering leaders, as well as UI designers, and a copywriter. I drove discovery, rapid prototyping, and testing through production release.
STATUS
Shipped, March 2024
CONTRIBUTION
Principal Product Designer
PRODUCT
iRobot Home (Mobile App, iOS & Android)
AUDIENCE
Robot Vacuum Cleaner Users, Premium Tier
SKILLS
UX Strategy & Design
Early Concepts & Storyboarding
User Research & Interviews
Rapid Prototyping & Testing
CONTEXT
iRobot users purchase Roombas to take cleaning off their to-do list and simplify the path to a clean home.
We knew from prior research that people tend to think about where they want to clean before how or what tool they’ll use: "I need to clean the kitchen" versus "I’m going to run my Roomba today."
The original iRobot app started as a single page, centered around Roomba control.
​
While our products handle the physical act of cleaning, this model left the user to decide where, when, and how to clean manually.​
As more of our new products came equipped with mapping capabilities and intelligent features, we saw an opportunity to leverage that intelligence to automate cleaning decisions for our users.
KEY INSIGHTS
Early discovery work uncovered insights about how we can better meet user expectations.
01
Match Users' Approach to Cleaning
Align with how users think about cleaning: where first, how second.
02
Offload the Mental “To-do” List
Minimize to-do lists, don’t just replace household tasks with device management.
03
Create a Frictionless Experience
Alert users when actions are required, don’t require users to check in with the app.
04
Build a Partnership with Users
Collaborate with users to achieve cleaning goals.
We decided to create an intelligent system, centered around spaces in the home, automating where, when and how to clean.
FOUNDATIONAL EXPLORATION
Early in the project, I created six storyboards exploring how an intelligent robot could work with users, from fully autonomous cleaning as part of smart home ecosystems to collaborative partners that suggest, but wait for approval.
In interviews, I asked users to rank the storyboards. Most were excited about autonomous robots, but most still wanted some level control. I was surprised at the lack of interest shown around integration with other devices and more complex autonomy. This provided some clarity: design for collaborative intelligence, but keep humans in the loop.
​
I moved to loose sketches to explore how this back and forth could work.
Designing Intelligence Users Trust
These early explorations left two parallel design challenges: How should we visualize cleanliness data? And how should we present intelligent recommendations?
​
I was excited to showcase our intelligence. I brainstormed with a designer on my team and we created some early designs that leaned into complexity: detailed heatmaps showing dirt distribution, numerical scores (even emojis!) for each room, explanations of the robot's reasoning.
For recommendations, I explored variations from suggestion to prominent defaults.
​
It felt impressive, and like we were giving users transparency into our technology and control over how it worked.
When I tested these designs with users, they were overwhelmed.
​
The visualization challenge: Users looking at heatmaps and numerical scores wanted to know which rooms to clean, not to parse through data. The precise information didn't help them make decisions - it got in the way. They had to work through complexity to find the simple answer they needed: which rooms need attention?
​
The interaction challenge: The visualization showed which rooms were dirty, but there was a disconnect between that and what "Smart Clean" would actually do. I experimented with ways to bridge the gap: a "See Plan" button, previews that auto-played upon pressing the recommendation. Users needed both visibility and control: even users who would ultimately accept our recommendations wanted to see the plan and have a chance to edit it before starting.
​
Both challenges had the same root cause: I was designing to showcase our intelligence, not to help users act on it.
THE APPROACH THAT WORKED
I stripped away everything users didn't need. I highlighted the main insight (which rooms need to be cleaned) explicitly above the map.
​
For the visualization: The numerical scores became a simple three-tier color system from dirty (dark green) to clean (light green). The heatmaps and detailed explanations disappeared entirely. The final design showed users their home's cleanliness at-a-glance.
​
For the interaction: I designed the system to suggest, not decide. The Dirt Detective algorithm predicts which rooms need cleaning and surfaces that as a recommendation: "Clean Dirty Rooms" using language that connects the plan directly to the map. It's presented as an option users can review and edit, not an automatic action.
WHY IT MATTERED
​
Simplification worked. In testing, users immediately understood the color-coded maps and acted on recommendations without hesitation. This principle of showing the insight while hiding complexity became foundational to how we designed intelligent features going forward.
​​
WHAT I LEARNED
​
Intelligence isn't valued for its sophistication, it's valued for the action it enables. Users didn't need to see our advanced capabilities, they just needed to know what to do next. The best design decision here was trusting our own intelligence enough to hide it.
SOLUTION
We delivered a user-centered experience that matches how people actually think about cleaning while leveraging our intelligence to automate cleaning routines.
INTRODUCING: DIRT DETECTIVE​
​
Cleanliness predictions, powered by the Dirt Detective algorithm, introduced the ability for Roombas to automatically decide if rooms needed to be cleaned, and prioritize them based on cleanliness.
Cleaning, Simplified
Users monitor their home's cleanliness through the My Home tab. Rooms are color-coded by dirt level, making it immediately scannable. They can clean recommended rooms with one tap, or create recurring schedules that automatically maintain their home.
Tailored to Your Home
By asking users about factors that affect cleanliness - like pets, kids, or high-traffic areas, we trained the algorithm to more accurately predict cleaning needs for each specific home.
​
This improved prediction accuracy over time while keeping the interface simple: users answer a few questions once, and the robot gets smarter about their home.
BEYOND LAUNCH
As we moved into 2025 we continued focusing our efforts on bringing users a space-centered, intelligent app purpose built to simplify their path to a clean home.
Smart Cleaning Settings
In Q1 2025, we released Smart Cleaning Settings to automatically choose options like vacuum suction level, mopping strength and number of passes based on predicted cleanliness of the space. These settings could be paired with "Clean Dirty Rooms", or could be applied to any custom cleaning routine.
Smart Notifications
Situations come up where we need users’ help to ensure we can keep their homes clean.
We have designed a out-of-app notifications strategy to provide clear and actionable requests for assistance when needed, further building a partnership with our users.
(No release date set.)
REFLECTIONS
This project pioneered the shift from robot-as-appliance to intelligent home partner, a paradigm that influenced future product development and established design patterns for more intelligent generations of robot.
01
Early Concepts Create Alignment
Testing storyboards and sketches, before any app screens were built or even designed, gave product, engineering, and robotics a shared framework to make decisions, and users something to react to. It became a core part of how I approached the 2025 redesign.
02
Trust Through Simplicity
My early designs showcased our advanced capabilities with numerical scores, obstacle detection and prediction explanations. Users didn't want to see the intelligence; they wanted to benefit from it. The final design kept only what they needed: which rooms to clean.
03
Partnership Over Automation
Users wanted intelligent recommendations, but only if they controlled when to act on them. The key was designing for partnership: the robot does the thinking, the user makes the call.