Sota: a tabletop robot built to talk with other robots, tried at reception desks and care homes

A 28 cm tabletop robot made to talk with other robots

Sota is a tabletop conversation robot developed by a research team under Osaka University professor Hiroshi Ishiguro and by Vstone. The Japan Science and Technology Agency (JST) is also named as a developer, and the character was designed by robot builder Tomotaka Takahashi. It was unveiled in January 2015 with its sibling robot CommU. The pair work by including a person in a conversation between several robots, so that the person feels part of the exchange. The name is short for “social talker” and the robot stands about 28 cm tall. Reservations for the developer edition opened on 28 July 2015 and ran until 21 August, at ¥99,900 including tax. It came in an Edison version and a Raspberry Pi version, and speech recognition, speech synthesis and dialogue control were to be added in later updates.

To a care home, a credit union and an office reception

Field trials began before the product shipped. In March 2015 NTT DATA tested Sota for safety checks, fall prevention and medication reminders at a special nursing home in Nishitokyo, Tokyo, from 25 March to 29 May, about two months. In July of that year NTT DATA, NTT and Vstone began a joint trial that links sensors and robots through the cloud. NTT supplied device-linking control and speech and dialogue technology, and Vstone supplied the robot. On 20 July 2016 a trial began in the lobby of Iwata Shinkin Bank’s head office in Shizuoka, where Sota explained account opening and products to customers; it was scheduled to run until 30 September. On 29 July a guidance trial linked to the reception system started at NTT DATA’s Toyosu Center Building. In 2019 Unicast added features to its robot reception system “Corporate Reception for Sota”, which uses the robot.

Trials took place, but we found no reported results

We found no published figures from these trials. The Toyosu announcement said only that the company planned to evaluate how well the setup suited reception work. Ishiguro said speech recognition had always been hard for robots. He added that people should adjust to what robots can do, and urged buyers not to stop at two robots but to buy three or four. Vstone expected to sell 3,000 units in the first year, mostly to businesses, with display-attention uses as an example.

Speech recognition sits in a separate cloud license

Vstone’s product page lists reception and guidance, presentations, remote conversation, elderly-care support, and research and app development as uses. The developer edition starts at ¥100,000 excluding tax, and the RoboPlat edition is ¥145,000 for the robot plus about ¥3,000 a month for 12 months. Speech recognition and synthesis come from Sota Cloud, whose one-year license is sold separately for ¥48,000. Converting a developer unit into the presentation edition removes speech recognition, the page says. The robot’s body shows the conversation, but the part that understands people’s speech lives outside it, and at a reception desk the reception system decides when the conversation starts. We could not confirm what the robot does once a license lapses.

If a robot cannot follow everything people say, what has to be designed for the exchange to still feel like a conversation?

What we haven’t confirmed

• How many units were actually sold (3,000 in the first year was Vstone’s forecast)
• The results of the care-home, Iwata Shinkin Bank and Toyosu trials, and whether any led to adoption
• Whether NTT took part in development beyond the joint trial. Robohub reported a development partnership, but NTT DATA’s announcement describes only a joint trial
• What the robot can do after a Sota Cloud license ends