Whisker's Litter-Robot 5 Pro AI features fail in testing

The Verge tested Whisker's Litter-Robot 5 Pro ($899) alongside its cheaper sibling, the Litter-Robot Evo ($599), for six months. Both models launched late last year and use the same core mechanism as earlier Litter-Robots: a cat enters a globe lined with litter, sensors detect it, and once the cat leaves, the robot rotates the drum to sift waste into a sealed bin below, so the owner only has to empty a drawer about once a week. That base scooping function worked well on both units. The 5 Pro's premium over the standard $699 Litter-Robot 4 is a dual camera system, sensors marketed as able to detect waste type, and AI features built on top: a system called WasteID, meant to tell #1 from #2 and speed disposal of the latter to cut odor, and a facial-recognition feature called Cat ID, built around a setup process named PawPrint, meant to track up to five individual cats.

Neither AI feature worked reliably. WasteID repeatedly failed to identify the type of waste and often could not report the weight of individual deposits. Cat ID could not tell apart the reviewer's two cats, a gray-and-white tabby named Smokey and a smaller tuxedo cat named Boone, and eventually began merging their records. The system recorded Smokey, who actually weighs 13.4 pounds, as having lost 2 pounds overnight because it mistook him for his lighter brother; the error was not self-corrected and both cats were still showing the same weight a week later. Fixing it required manually resetting each cat's weight profile, reassigning misidentified visits, and correcting video logs.

Whisker offered several explanations and fixes over the course of the test. The company said the unit's feces-sensing gas sensor was off and corrected it remotely; when that did not resolve the problems, Whisker determined the unit's scale itself was faulty, which the company said could account for both the WasteID and PawPrint errors, and said fixing that would require a new machine. On facial recognition, Whisker said poor lighting and weak initial images can throw off the camera, that the system falls back on other signals such as the onboard scale when uncertain, and that it has "improved the enrollment process so it's more selective about the images it uses for facial recognition." Whisker also said the interior camera's poop-view recording, which the marketing implies is already available, is coming soon; today only the front-facing camera records visits, while the interior cameras offer just a live view.

The 5 Pro's AI features run on top of the $8-a-month Whisker Plus subscription, which the reviewer said delivered no actionable or useful health insight during the test period. A newer feature called Patterns, which summarizes each cat's day (visit counts, time spent, weight), rolled out several months after launch and moved partway toward that goal but still gives no detail on waste type; in practice, 5 Pro owners were paying the monthly fee for months before Patterns arrived for features that were not working.

The reviewer's recommendation: skip the 5 Pro's AI premium. Buying the standard Litter-Robot 5 instead saves $100 by dropping the camera while keeping WasteID and the rest of the features. For anyone who does not need WasteID at all, the non-AI $699 Litter-Robot 4, which supports up to four cats, is recommended over the cheaper $599 Evo, which has a smaller bin, no litter-level monitoring, and felt less sturdy in testing. Whisker separately sells a $29 RFID PetTag to help the system tell similar-looking cats apart. The reviewer's broader conclusion is that Whisker, like other smart-home makers, overestimated what its AI could deliver once deployed in real homes rather than a lab, where conditions are far less predictable.

Key facts

  • Whisker's $899 Litter-Robot 5 Pro AI features, WasteID waste-type detection and Cat ID facial recognition, failed to work reliably over six months of testing alongside the $599 Litter-Robot Evo.
  • The facial-recognition system could not distinguish the reviewer's two cats and merged their records, recording 13.4-pound Smokey's weight as his brother Boone's, an apparent 2-pound overnight loss that persisted for a week uncorrected.
  • WasteID repeatedly failed to identify waste type and often couldn't report individual deposit weights, despite the $8-a-month Whisker Plus subscription that is supposed to cover these AI health features.
  • Whisker attributed the issues to a faulty gas sensor and then a faulty scale on the reviewer's unit, said facial recognition is 'built to learn,' and said full correction of the scale issue would require a new machine.
  • The reviewer recommends the $100-cheaper standard Litter-Robot 5, which keeps WasteID without the camera, or the AI-free $699 Litter-Robot 4, over paying the 5 Pro's AI premium.

Why it matters

This is a concrete, hands-on case of a consumer AI feature failing once deployed in an unpredictable home environment rather than a lab, from a company charging a recurring subscription for that feature. It illustrates a pattern the reviewer says is common among smart-home makers: shipping AI capability ahead of it actually working, then asking buyers to pay monthly for features still in progress.

Who it affects

Buyers of the $899 Litter-Robot 5 Pro, particularly multi-cat households paying the $8-a-month Whisker Plus subscription for AI health features that did not deliver useful insight during the test. It also matters to anyone comparing the Evo, the standard Litter-Robot 5, and the non-AI Litter-Robot 4, and more broadly to shoppers deciding whether to pay a premium for AI on newly launched smart-home hardware.

How to use it

The reviewer's guidance: skip the 5 Pro's AI premium. The standard Litter-Robot 5 costs $100 less, drops the camera, but keeps WasteID and the other core features. Anyone who doesn't need WasteID at all is better served by the non-AI $699 Litter-Robot 4, which supports up to four cats, over the cheaper $599 Evo. A $29 RFID PetTag is sold separately to help the system distinguish similar-looking cats. Buyers of the 5 Pro should expect to manually correct cat weight profiles and misidentified visits while the system is learning.

How solid is it

The account comes from a first-hand, six-month test of two Litter-Robot models run side by side by the outlet. Whisker responded directly to the reviewer's reported problems, attempting remote fixes (the gas sensor) and diagnosing a hardware cause (a faulty scale) for the rest, and provided on-record statements quoted in the piece, including on the facial-recognition enrollment process and the still-missing interior-camera recording.

Risks and caveats

The findings rest on one review unit and one reviewer's two cats rather than a broad sample, and Whisker attributed part of the failure to hardware faults specific to that unit (the gas sensor, the scale), so other units may perform differently. Whisker says the facial-recognition system is built to learn and should improve, and a partial fix (the Patterns feature) had already shipped by the time of the review, so the described failures are not necessarily permanent.

“improved the enrollment process so it's more selective about the images it uses for facial recognition.”

— Whisker, in a statement to The Verge