Roomba users no longer need to manually edit their home maps: new Roombas could do it automatically, simplifying users' paths to map-based features.
Unblocking Roomba Users with Auto Customized Maps
PROJECT OVERVIEW
In 2023, iRobot launched an initiative to automatically learn key attributes of users' homes, starting with the Roomba's ability to identify room types.
​
Auto-customized maps dramatically improved user onboarding: 97% of users understood their map after one run, and gained access to fully customized maps 7x faster.
​
I led design for Auto Room Naming on a technical team of robotics and cloud engineers, driving the feature from user testing through production release. I delivered comprehensive UX flows and behavioral specifications, fully vetted for feasibility and accounting for diverse home layouts and edge cases.
STATUS
Shipped, March 2023
CONTRIBUTION
Senior Product Designer
PRODUCT
iRobot Home (Mobile App, iOS & Android)
AUDIENCE
Robot Vacuum Cleaner Users, Premium Tier
SKILLS
User Research & Interviews
Rapid Prototyping & Testing
UX & UI Design
CONTEXT
​
​​
A cumbersome map customization flow was blocking Roomba users from key map-based features. As a result, only 20% were engaging with these features because getting an accurate home map was complex and time consuming.
​
This was problematic since we know users approach cleaning by space, and our experience was becoming increasingly centered around the home map.
​
In 2023, new Roombas came equipped with the ability to identify room types. This meant we could offer users automatically customized home maps, unlocking map-based features early, and with little to no user input.
DISCOVERY
We conducted a 4-week diary study with 8 participants who recorded their experience with auto-customized maps using the current (mandatory editing) flow.
WEEK 1
1st mission + map customization with new algorithm
WEEK 2
Interview: Map Viewing + Feedback
WEEK 3
Record observations and thoughts, make map edits
WEEK 4
Use Robot as desired, final thoughts + exit survey
WHAT WE LEARNED
01
Room labels were critical for map comprehension
Room labels turned abstract shapes into comprehensible maps. Without them, users struggled to orient themselves.
02
The mandatory editing flow created unnecessary friction
Users who needed edits made them naturally. Forcing happy users through editing just delayed their first clean.
03
Zone creation was premature in initial setup
Users couldn't visualize zones before understanding their space. Most skipped over this step.
Finding the Confidence Threshold
The new 2023 Roombas could identify room types with varying levels of confidence. Engineering wanted a high confidence threshold which meant only auto-naming rooms we were very sure about. Their concern was valid: mislabeling rooms would break user trust.
​
But I was concerned about the opposite problem: unnamed rooms. Without room names as wayfinders, users couldn't orient themselves on our abstract maps. Especially with our the list-based cleaning interface, "Room 1, Room 2, Room 3" all looked identical, making cleaning with the app nearly impossible.
The algorithm could identify rooms at two levels: room types (bedroom, kitchen, bathroom) and specific room names within those types (master bedroom, guest bedroom).
Engineering's proposal was straightforward: show whatever had very high confidence, whether that was a specific name, a generic type, or nothing at all.
​
But this created an inconsistencies for the user. One map might show "Master Bedroom, Bedroom 1, Room 2, Kitchen, Room 4" which felt like a chaotic mix of specific names, generic types, and unnamed rooms. The benefit of this proposal is anything that was named, was very likely to be named accurately. However, making the threshold so high was at the cost of maps making sense to the humans on the other side.
​
This wasn't just about functionality - it was about comprehension. In 2023, our maps were abstract shapes without walls. Room labels were the only wayfinders we could offer from the get go. A lack luster or inconsistent labeling scheme would confuse users and break trust, possibly more than an inaccuracy here or there.
THE APPROACH THAT WORKED
I proposed a two-part approach: always expose room types, never specific names, and lower the confidence threshold for room types slightly, to ensure most rooms got labeled.
​
This addressed both problems. Consistent labeling by room type ("Bedroom 1, Bedroom 2, Kitchen") kept maps feeling consistent, and offered better accuracy. Better room coverage meant fewer unnamed rooms that didn't help with wayfinding.
​
Users could orient themselves on abstract maps and easily rename rooms if specificity mattered. Good enough accuracy with easy correction was better than near-perfect accuracy.
WHY IT MATTERED
​
Alpha testing validated the approach: 4.09/5 stars for label accuracy. Users oriented themselves using consistent room type labels and simply renamed rooms when they wanted more specificity. Ultimately, in production, we got users to customized maps 7x faster, unlocking map-based features early.
​​
WHAT I LEARNED
​
"Good enough" automation that solves most of the problem beats perfect accuracy that leaves too much unresolved.
Making Editing Optional, but Obvious
The diary study showed that we didn't need to force map editing. I redesigned the flow so users could skip directly to cleaning if they were happy with their map, while making sure editing was discoverable for those who needed it.
​
Below, you can see the original map editing flow, with mandatory steps side by side with the new flow, offering optional editing with removed zone creation.
This update preserved simplicity and speed for users with accurate maps while supporting those who needed changes. Users could add zones later, after they'd actually used their robot and understood their home layout.
INTRODUCING: AUTO ROOM NAMING
​
Automatically identifies rooms and labels them for you after just one run.
Quick context to set users up for success
Before presenting the first home map, we added an educational screen to provide some basic map comprehension support and context for the upcoming map review flow.
Streamlining the happy path
For users who were happy with their initial map, no editing options were necessary at all.
Simplified map editing
For users who were not happy with their initial map, editing options were made very simple.
We removed zone adding & editing all together, pushing those prompts later in the user journey.
"Great work. Best mapping part of any robotic device I've used to date. Thorough and easy to use."
ALPHA TESTERS SAID...
IMPACT
6 months post-launch, users understood maps quicker, and accessed customization 7x faster with high satisfaction.
OVERALL
EXPERIENCE
4.33
out of 5
MAP
COMPREHENSION
97%
of users understood
their map immediately
after mapping run
+62%
of users had access to map-based features early in their journey
MAPS CUSTOMIZED WITHIN 5 RUNS
USERS GAINED ACCESS TO CUSTOMIZED MAPS
7x faster
AND with no input
from the user
BEYOND LAUNCH
As we moved forward with a more a more intelligent, home-centric app, easy access to a customized map has become key to the Roomba experience.
Forced Mapping Run
In early 2024, we released forced mapping, which makes map review mandatory to even enter the app experience.
This feature solidifies maps as an essential piece of the overall Roomba experience.
Continual Map Improvements
In Q1 2025, we released improved maps with automatic detection of furniture, floor type and other wayfinders in the home to work towards this goal.
REFLECTIONS & LEARNINGS
This project showed me how the best solutions can emerge from validating different perspectives, not choosing between them. Engineering feared mislabeled rooms would break trust. I worried unnamed rooms would confuse users. The balance was finding a third path with consistent room types and reasonable (not perfect!) accuracy, that addressed both concerns.
​
Leading with questions helped: "What's the actual confidence distribution?" let us see the data together. "What's worse, mislabeled or unnamed?" reframed it as a shared problem. The diary study gave us evidence we needed to make a decision: users tolerated generic labels and corrected what they needed. When technical constraints and user needs feel like competing forces, the answer is usually a third option neither side has articulated yet.