B2C
AI Smart Home
Consumer Mobile

AI-powered Autonomous Home Cleaning System

Reworking the DEEBOT X2 from a robot you operate into one you trust: one-tap AI cleaning, a map that draws itself, and pet-safe navigation.

Overview

An AI auto-cleaning experience that replaces complex manual setup with one-tap intelligent control.

Note

ECOVACS positioned the DEEBOT X2 as fully autonomous, but rising customer complaints told another story. The hardware was capable; the experience still asked people to edit maps, choose between a dozen settings, and hope it avoided the worst messes. I led the design that closed the gap between what the robot could do and how intelligent it actually felt.

Role: Product Designer
Team: Product Design
Software Eng
Machine Learning
Hardware R&D
Product Mgmt
Skills: Product Strategy
Research Synthesis
Interaction Design
System Logic
Prototyping
Timeline: May to Aug 2024
12 weeks

What I owned

Defined success metrics and guardrails, and aligned five teams on one product contract.
Turned usability findings into a prioritized roadmap, MVP scope, and acceptance criteria.
Partnered with Robotics, ML, and Eng so the UI matched real robot constraints, model confidence, and failure modes.
Shipped an AI hosting system that cut configuration without sacrificing safety or cleaning quality.

Core problems

Autonomy that still felt manual.

The X2 was packed with capability, but TikTok and e-commerce reviews told another story: people found it complicated and, at worst, untrustworthy. We combined complaint analysis, App Store review mining, and interviews with pet owners and first-timers. Three pain points came up again and again.

First impression

Setup was a chore

Before the first clean, users divided rooms by hand, drew virtual walls, and walked a long setup. A rough first impression that hit almost everyone.

Everyday friction

Too many knobs

Suction, water flow, cleaning passes: more than ten parameters, with no clear way to choose. The product felt complicated instead of smart.

Trust-breaker

Pet-waste accidents

Sometimes the robot failed to recognize pet waste and smeared it across the house. Rare, but catastrophic. It broke trust instantly.

In their words

“I thought it was automatic, but I still had to redraw the map myself. It feels more manual than smart.”

Robert Rose

“There are too many cleaning settings. I don't know which one to choose.”

Alex Smith

“If it runs over pet waste, that's a disaster. I'd completely lose trust.”

Jessica Davis

Goals

Make autonomy feel effortless, and earn trust at the riskiest moments.

The X2 did not need more features. It needed fewer decisions and more confidence. Three goals framed the work:

Make mapping feel truly automatic

Auto-mapping existed, but users still split rooms, renamed spaces, and fixed boundaries. Cut those corrections so mapping feels confident on its own, especially for first-timers who expect plug-and-play.

Take decisions off the user

Instead of asking people to configure the robot, shift the call to the system: let AI read floor types and usage patterns, then recommend or apply the right settings.

Design for trust in high-risk moments

Pet-waste failures destroyed trust. Prioritize high-risk detection and clear system feedback so the robot feels dependable, not unpredictable.

Impact

The autonomy finally felt like autonomy.

After launch, the experience told a different story than the complaints that started it. Setup got out of the way, the AI mode became the default behavior, and trust returned.

Setup: Setup got out of the way; One-tap AI cleaning replaced manual map editing, room-splitting, and the long first-run setup.
Adoption: AI cleaning became the default; People let the system read the room and pick the strategy, instead of tuning a dozen settings by hand.
Trust: Trust came back; Pet-waste, the scariest failure, became a non-event, and feedback shifted from “too complex” to “feels smart.”

Design Challenge 1

How might we make setup something that just happens, not a chore?

Now the robot generates the map, divides rooms, and labels each room type on its own. People only review and make small tweaks. Setup goes from a task to a glance.

A contextual first-run prompt gives simple prep tips, then guides users in when no map exists yet.
Mapping runs long, so interruptions are explicit: early-exit warnings prevent accidental data loss.
After the scan, people name the floor and adjust labels only for the rare correction.

Mapping · before → after

Map editor — splitting rooms and drawing virtual walls by hand — BeforeEditing the map by hand

AfterThe map draws and labels itself

Design Challenge 2

How might we replace a dozen settings with a single, trustworthy tap?

Tap Start, and the system reads each room's type and floor material to pick the strategy itself: suction, water, passes. The experience shifts from manual control to something that behaves like an agent working on your behalf.

AI hosting sits at the whole-home cleaning entry, visible and one tap away, not buried in settings.
Advanced parameters move to a separate custom tab, so power users keep full control.
A first-time explainer sets expectations, a live status shows AI is driving, and a post-clean report makes every decision traceable.

AI Cleaning · before → after

A dozen manual cleaning settings — mode, water level, passes, efficiency — BeforeA dozen settings, set by hand

AfterOne tap, AI picks the strategy

ThenThen hands back a report

Design Challenge 3

How might we make the scariest failure a non-event?

We retrained the recognition system to reliably detect pet waste and steer around it mid-clean, turning the worst failure mode into a non-event.

Pet-Safe · before → after

Pet waste smeared in long streaks across the living-room rug — BeforeRan over pet waste, smeared across the house

The robot's underside and roller brush caked with pet waste — BeforeRan over pet waste, smeared across the house

AfterDetects waste and avoids it

An honest trade-off

In build, a real constraint surfaced: pet-waste detection and fine-particle cleaning could not run at the same time. Rather than silently pick one, we shipped a clear toggle with an explanation, so people make an informed choice for their own home.

Takeaways

What this taught me.

Start with real household behavior

People do not think in cleaning modes. They think in outcomes: make my home clean, and do not bother me. The value came from removing decisions, not adding options.

Design for trust, not just automation

Autonomy only counts when people believe in it. One catastrophic failure outweighs a hundred smooth runs, so high-risk moments deserve the most care.

AI should feel invisible but dependable

The win was not more controls, it was fewer. The system makes the call, stays transparent about it, and earns the right to be trusted by default.

More & next

More that rounded out the X2.

Beyond the three headline features, a few more refinements rounded out the X2 experience.

Yiko · Voice

Just say it out loud.

We refined Yiko, the on-device voice assistant, so people can start a clean, send the robot to a specific room, or pause it, all hands-free without ever opening the app.

Onboarding

Guided custom cleaning.

Guided onboarding helps users learn how to add custom cleaning areas, select zones on the map, and turn on AI-assisted cleaning. Contextual instructions sit right in the interface, so advanced features feel approachable for new users.

Scheduling

Smarter scheduling across modes.

The scheduling flow now supports full-home, custom, and zone-based cleaning with repeat options and reusable presets. People can plan routines in advance and tailor tasks to different rooms, floors, and household scenarios.

Next steps

Where this goes next.

After launch, we extended the Yiko settings architecture to support growing AI capabilities: voice control with a reviewable interaction history, intelligent error detection that replaces error codes with visual fixes, and smart consumables reminders.