Japan's AIREC and the Slow Work of Trusting a Machine

There has been plenty of talk and hype about robots with plenty of videos of them running, jumping and even dancing. However, there is not much discussion about what real life problems they are aimed at solving

A bit of a background, before we delve into technicalities. As you may well be aware, Japan’s population is aging while the number of able people who work in age care is declining. The projected estimate is a deficit of 570,000 by 2040. To make things worse, the care workers in Japan are among the lowest paid, which contributes to a very high turnover rate.

This is where AIREC ( AI-driven Robot for Embrace and Care) comes in, a robot prototype developed at Waseda University led by Professor Sugano Shigeki. The project is a part of the government-funded Moonshot R&D Program, the goal of which is to assist agecare workers during the physically demanding tasks. This in turn is to free them to do what we humans do best - provide empathy and communication.

What does it look like

It is 170cm tall (an average of a Japanese male), weighs 150kg and uses several sensors, such as fish eye and depth cameras, pressure-sensitive fingers and biological data sensors that collect vital signs such as heart rate, blood pressure and body temperature.

What it can do

As of 2026 it is capable of patient repositioning, bed-exit support, can change socks and, a world’s first, able to perform ultrasonic examination.

Beyond that, it showed the capability of cooking scrambled eggs and cleaning the tabletop. It cannot yet string up a series of tasks independently.

The AI stack

Very early on, the engineers devised the AI stack that uses what they call deep predictive learning - a framework where the robot learns various skills by observing human demonstrations. It can then predict the future states to guide the actions.

They also heavily utilise NVIDIA’s Isaac Sim framework for simulation based training tasks, such as force estimation between objects.

Challenges and potential answers

Cost. At ¥10 million (USD 67,000) per unit, the robot is cost-prohibitive for anything but large organisations with huge budgets. Good news is that prices fall as the adoption grows.

Reliability. Not quite there yet, where those who tried it, reported malfunctioning at various stages. This is normal with any new technology. Early radios, TVs and Computers required a team of engineers to maintain them.

Complexity of subjects. Humans come in all sorts of shapes and sizes, where it is downright impossible to program every single type into the robot’s behaviour model. Good news is that we no longer need to train models manually but use various learning types, including simulation.

The biggest potential showstopper of them all however is social acceptability. The project includes the ongoing ELSI (Ethical, Legal and Social Issues) research to address public acceptance challenges. On a personal level however, we cannot simply say to a person “trust this robot with your life” and not be told off.

The current phase is running between 2026 and 2030 by which the commercial deployment of the robot is projected to begin, with full autonomy targeted by 2050.

Whether public trust can move as fast as the engineering does remains to be seen.

Citations:

https://www.trade.gov/market-intelligence/japan-healthcare-caregiving-technologies

https://www.ilpnetwork.org/presentation/shortages-of-care-workers-and-their-training-and-retaining-in-japan/

https://airec-waseda.jp/en/about_en/

https://blogs.nvidia.com/blog/japan-science-technology-agency-develops-moonshot-robot/

https://helloworldjapan.substack.com/p/robotics-in-care-how-japan-is-using

https://sinolytics.de/global-business-news/blog/geolytics/robots-elderly-care-lessons-from-japan/

https://www.jst.go.jp/moonshot/sympo/20210327/pdf/MS3_SuganoPM.pdf