本文へジャンプ

成果報告書詳細
管理番号20170000000656
タイトル*平成28年度中間年報 次世代人工知能・ロボット中核技術開発 革新的ロボット要素技術分野 人共存環境で活動するロボットのためのHRI行動シミュレーション技術
公開日2017/8/30
報告書年度2016 - 2016
委託先名株式会社国際電気通信基礎技術研究所
プロジェクト番号P15009
部署名ロボット・AI部
和文要約
英文要約In this project, we are conducting research about innovative robot integration technology for next-generation robots. Due to the complexity of human beings, it is difficult to foresee what is going to happen when a robot is put in a real environment to interact with people. This issue hinders the development of sophisticated robots. To tackle this problem, we are developing an HRI (human-robot interaction) behaviour simulator for assisting the process of system integration. This simulator could greatly improve the efficiency of development, by reducing the time spent on costly trial-and-error sessions in real-field.

For this purpose, the following works have been conducted.

1. Research and development about modelling people’s human-robot interaction behaviour
We carried on the work started the previous fiscal year, where we developed a basic model of pedestrian that comprises a social force model and a model of behaviour around the robot, and undertook improvements and integration of the human-robot interaction model of pedestrians.

2. Research and development about reproduction of phenomenon which occur in real-field conditions
In this fiscal year, to evaluate the realism of the reproduction of real environments in the simulator, 40 minutes of data with the robot running in a shopping mall were prepared, and a calibration method using a genetic algorithm was also developed to make the simulation similar to the observed reality.
For the sake of reproducibility of various types of case-specific situations, a few ways of customizing the simulation relative to the user’s specific needs were developed. The possibility to define scenarios, where events can be described through time in a simulation, was developed. Custom people behaviour can also be defined.

3. Evaluation of simulation quality
The quality of the reproduction was evaluated. For this purpose, a simple interaction robot that moves around an area and greets people was prepared. Using data from this robot, the simulator was calibrated and the quality of reproduction could be evaluated. To evaluate the quality and calibrate the parameters, the following indicators were computed: pedestrian velocity (average speed of all humans in the area), density far from the robot (amount of people more than 6 meters away from the robot), density around the robot (people between 2 and 6 meters away from the robot), density close to the robot (people closer than 2 meters from the robot), interaction count (amount of people who stopped in an area of 6 meters around the robot). Another indicator, robot velocity (average speed of the robot over the session. As the robot slows down or stops to avoid collisions with people in front of it, this velocity is strongly influenced by the people’s interacting behaviour) was used for performance evaluation only. By taking these indicators’ average values from about 500 simulations and comparing them with the values from real data and checking the difference relative to the standard deviation of each indicator, we could check whether the simulation quality is satisfying or not. The study showed that the calibration can adjust the simulation parameters so all the indicators computed from the simulation fit with a value close enough from the real data (that is, within standard deviation). However, without the HRI model we developed, the robot velocity indicator does not reach a value close to the real data. This showed that without a human interaction model, the robot’s behaviour will not reflect reality and proved that a human-robot interaction model is essential in a simulator for social robots.

4. Research and development about robot development using a simulator, use in integration work

Studies were conducted to investigate the situations where a simulator could be helpful and increase the efficiency of robot development.
The developed simulator was also used for integration work.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る