PARO: a therapeutic seal robot still on sale after 20 years

Why a baby harp seal, not a dog or a cat

PARO was developed by Takanori Shibata at Japan’s National Institute of Advanced Industrial Science and Technology (AIST), who began the research in 1993. He first experimented with dog and cat robots. People compared them with real animals and were disappointed, so he chose an animal few people know: a baby harp seal. He explained that the less familiar people are with an animal, the more easily they accept a robot of it. PARO is 57 cm long and weighs about 2.55 kg. Sensors for touch, light, sound, temperature and posture let it recognise stroking, being picked up and its own name, and it moves its head, flippers, tail and eyelids. It has no screen, app or account.

From Guinness to the FDA: becoming a therapeutic tool

PARO was made for hospitals and care homes where animals are hard to bring in. Guinness recognised it as the world’s most therapeutic robot in 2002, and leasing to care facilities began in September 2004. In 2009 the US FDA cleared it as a medical device. Denmark distributed PARO to nursing homes with public funds after a 2008 pilot and in 2010 ran a training course for care staff. In Japan it is a listed device for dementia care support in the national welfare equipment database. Institutions gained a reason to buy it, and staff a way to learn how to use it.

It helps, but not much more than a plush toy

PARO is the most studied of all robotic animals. In a 2017 randomised trial with 415 people in 28 care facilities in Australia, the PARO group showed more engagement and pleasure and less observed agitation. On the standard agitation scale, however, there was no difference. A cost analysis of the same trial found that a plush toy reduced agitation scores at slightly lower cost. A review of studies in care settings listed upfront cost, difficulty cleaning the fur, maintenance costs, extra work for staff and concern about treating adults like children as barriers to use.

A robot that promised only what it could do

PARO never pretended to hold a conversation. What it promised and what it could do were nearly the same. Because it does not rely on company servers, the model released in 2013 is still on sale today. Jibo, which lost its functions when its maker closed, is the contrast. It is also possible that PARO lasted less because of the robot itself than because of certification, procurement and staff training.

How far apart are what this robot promises and what it can actually do?

What we haven’t confirmed

• Total units sold as of 2026, and the share by country
• Cases of facilities that stopped using PARO, and why
• The FDA listing number and the original Medicare coverage document
• Whether staying offline is a stated design principle