2019 Fiscal Year Final Research Report
Study on behavior learning for a single-seat personal mobility vehicle capable of traversing rough terrain
| Project/Area Number |
17K00364
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent robotics
|
|---|
| Research Institution | Wakayama University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2017-04-01 – 2020-03-31
|
| Keywords | PU-GAN / LaserVAE / 深層強化学習A2C |
|---|
| Outline of Final Research Achievements |
In this research project, we aimed to develop an automatic learning method for a single-seat personal mobility vehicle (PMV) to learn the action sequences (generation of combinations of wheel and leg actions) for various driving conditions such as flat roads, uneven terrain and several steps. To realize this, we confirmed that (1) the PMV's surrounding environment can be sensed by a 3D laser scanner (hereafter referred to as 3D Lidar), and more dense 3D data (point cloud data) can be generated from sparse 3D data (point cloud data) by using the PU-GAN method, and that (2) the simulated PMV can automatically learn the step override motion in a simulation environment with one step using the deep reinforcement learning algorithm (A2C).
|
| Free Research Field |
知能ロボット
|
| Academic Significance and Societal Importance of the Research Achievements |
PMVは，高齢者を含めた移動困難者のQOLを向上させる移動支援機器として普及が期待されているが，都市環境においても整備されていない段差は数多く存在する．本研究で開発した手法が実用化できれば，整備されていない環境中を踏破できる能力を持ったPMVの普及が進み，PMVの生産という新製造業を創出できると考えられ，社会的な意義が大きい．
本研究で開発した手法により，数段の階段を踏破できるような能力をPMVに与えることができ，高速な車輪モードから低速な脚動作モードへの滑らかな切り替わりなどが実現できるようになると考えられる．このようなソフトウェアを搭載したPMVは，類似の開発例がなく，学術的な意義も大きい．
|
