Guide Released for Safe AI Robot Deployment

The Robotics Subworking Group, part of the Business Demonstration Working Group of the AI Safety Institute, developed evaluation guidance to promote the safe deployment of AI robots in society.

The background is the growing importance of AI-driven perception, judgment, and behavior in service robots and collaborative robots, in addition to conventional control technologies. For transportation and delivery, robots are expected to do more than avoid obstacles and generate routes; they must also respond naturally to users, assess situations, and allow human intervention.

The guide evaluates issues that are difficult to capture through machine safety and functional safety alone, including not only AI misrecognition and misjudgment but also psychological impacts, the handling of data and logs, and explainability and verifiability.

Robot manufacturers, robot system integrators, and component manufacturers will find it easier to identify AI safety considerations during the development, design, provision, and operation of products and services.

For users, consideration of impacts such as collisions and contact, intimidating speech, behavior that causes anxiety or discomfort, discriminatory treatment, and the spread of false or misleading information will broaden attention to safety when people and robots share spaces.

Updates are planned to reflect changes in the field and international regulatory trends, and the guide is expected to serve as a foundation for safety-conscious deployment by businesses and the creation of sustainable business value.

The scope covers AI robots that use AI to share spaces with people while moving, interacting, transporting items, and performing similar tasks. The guide presents perspectives for identifying and evaluating risks that may newly arise from AI use and considering risk-reduction measures based on the results.

Risks are organized into physical risks, such as collisions and contact; psychological risks, such as intimidating speech and behavior that causes anxiety or discomfort; and social risks, such as discriminatory treatment and the spread of false or misleading information.

The evaluation perspectives that practitioners should check during design, development, provision, and operation were organized with reference to AISI's “Evaluation Perspectives Guide for AI Safety.” The assumed use cases are product delivery robots in restaurants and similar establishments and remotely operable small delivery robots.

Organizations participating in the SWG, led by the National Institute of Advanced Industrial Science and Technology, and the secretariat of Mitsubishi Research Institute were involved in developing the guide. Participating organizations included IHI, Kawasaki Heavy Industries, Panasonic Holdings, Hitachi, Fujitsu, and Mitsubishi Electric.