成果報告書詳細
管理番号20160000000585
タイトル*平成27年度中間年報 次世代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2016/7/22
報告書年度2015 - 2015
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1. Modelling of human-robot interactions
In the 2015 fiscal year we developed of a basic model of pedestrians. A well-known model for modelling the movement of pedestrians in real-world environments while avoiding collisions is the Social Force Model. In this model, the interactions between people are described using "social forces", which can accurately predict the pedestrian movements. However, the original model by Helbing and Molnar (Phy. Rev. E, 1995) was developed for describing high density panic and escape situations. Recently we developed a model which better reproduces the people's motion in shopping areas and other less crowded spaces (Zanlungo et al., PLOS ONE, 2012). This Extended Social Force Model is used as a basis for modelling the flow and interactions between pedestrians. Additionally, the model has been further extended to comprise the stopping behaviour of pedestrians, which can frequently be observed in shopping areas.
In addition to interactions between humans, there is also a need to model the movement of people in the vicinity of robots. The presence of a robot often affects the surrounding people, which show a range of different behaviours. In order to study the patterns of behaviour during these interactions, we let the robot offer guidance services in a populated space, and later analysed the collected video and person position data. The following distinctive human-robot interaction types were identified: approaching to interact; stopping first and later approaching; watching from distance; slowing down to look at robot; not interested. The frequency distributions of each pattern as well as their characteristics (duration of interaction, distance from robot, etc.) were also determined. This was integrated with the Social Force Model to be used as a basis for reproducing the human behaviour and human-robot interactions inside simulations.
2. Collection of environment data and model integration
Our goal is to closely reproduce the phenomena which happen in the real world, so in this fiscal year we collected data from a populated environment. We built a dataset that includes the shape of the environment and the behaviour of persons, both to be used in the simulations as well as preparation for the model tuning based on data to be studied in the work to follow.
For better efficiency, we reorganized and prepared a previously collected large dataset of person behaviour (23 days, with video and person position data). Since this dataset does not provide enough information on the shape of the environment, we collected additional data using a 3D laser scanner mounted on a moving robot. This allowed us to make a 3D map of the environment and create a 3D model for the use in simulations.
3. Development platform with integrated simulator
We have been working on a development platform for robot integration tasks which integrates a simulator. Several existing development environments for robots contain a physics simulator, 3D viewer, tools for describing the robot and simulating the sensors. We chose to base our platform on MORSE, an open-source, widely used simulator that supports many different robots and robotic middleware.
Although MORSE provides one simulation model of a human, it was meant to be used for teleoperation and not autonomous operation. We therefore added to the MORSE simulator a human model which implements the model of pedestrian movements and human-robot interactions that were described above. This way we are able to reproduce the autonomous movement of pedestrians inside an environment, including also the animation of the walking behaviour and 3D visualisation of the simulated world, which allows us to obtain simulations of increasingly higher levels of reality. Moreover, we created a library of 3D different human shapes (at the moment it contains 80 adults and 10 children) and their movements, which can be used inside the simulator.
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