日本ロボット学会誌
Online ISSN : 1884-7145
Print ISSN : 0289-1824
ISSN-L : 0289-1824
40 巻, 2 号
選択された号の論文の21件中1~21を表示しています
特集について
随想
論文
  • —写実的なCG画像で学習した深層畳み込みニューラルネットワークによるImage-to-Image変換—
    三上 泰史, 石川 弘也, 原 直裕, 夏迫 和也, 空閑 融, 大和田 恭平, 北浦 勇人, 上田 隆一
    2022 年 40 巻 2 号 p. 143-153
    発行日: 2022年
    公開日: 2022/03/17
    ジャーナル フリー

    This paper proposes a method for training deep convolutional neural networks (DCNNs) as a detector of plant stems partly hidden by leaves. The detector is assumed to be used by an agricultural robot that treats near trees in dense fruit vegitable fields, where the stems of trees are hidden not only by leaves, but also by the totally green confusing background. We tackle this difficult problem by training DCNNs with datasets that are compose of realistic computer graphic images (CG images) and annotated images of foreground leaves and stems. Physically-based Rendering (PBR) and Image-based Learning (IBL) are utilized for creating the CG images that make DCNNs distinguish foreground trees from the background. We have trained two DCNNs with the CG images and created a main stem detector with the DCNNs. In the experiments, the detector has distinguished near trees from the confusing background and has found a main stem partly hidden by leaves.

  • 関口 叡範, 武居 直行
    2022 年 40 巻 2 号 p. 154-161
    発行日: 2022年
    公開日: 2022/03/17
    ジャーナル フリー

    This paper discusses motion planning techniques for redundant articulated robots. Motion planning in this paper means to generate joint trajectories that can avoid obstacles. In the proposed motion planning method, candidate waypoints are generated from the overall shape of the initial and goal configurations. All procedures of the proposed method are constructed only by computing inverse kinematics, and it does not require path planning in joint space. The effectiveness of the proposed method is verified through performance evaluation test using a 7-DOF manipulator. In the performance evaluation test, not only the trajectory length and calculation time, but also the risk of collision with obstacles that are not considered in the simulator are evaluated.

  • 星野 智史, 吉田 遊友
    2022 年 40 巻 2 号 p. 162-165
    発行日: 2022年
    公開日: 2022/03/17
    ジャーナル フリー

    Motion planning is an essential capability for autonomous navigation of a mobile robot. This paper focuses on end-to-end motion planners. The presented motion planner predicts a control output from the image input. For input-output mapping, the planning policy is represented by a CNN. In this paper, both the static and dynamic obstacles are the targets. For dynamic obstacles, the robot is required to take into account the time series variation in the input images. Therefore, we propose to use a LSTM block within the CNN. Imitation learning is applied to the planning policy for behavior cloning. The results of experiments show the effectiveness of the motion planner through mediated perception not only for a standing person, but also for a walking person as dynamic obstacles.

  • 境 辰也, 堀井 隆斗, 長井 隆行
    2022 年 40 巻 2 号 p. 166-169
    発行日: 2022年
    公開日: 2022/03/17
    ジャーナル フリー

    To realize advanced interaction between autonomous robots and users, it is important for robots to aware the difference in their state space representations (i.e., world models). As a first step toward this goal, we propose a method to estimate user's world model based on queries. In our method, the agent learns distributed representation of world models by graph2vec and generates concept activation vectors (CAVs) that represent the meaning of queries in latent space. The experimental results show that our method can estimate user's world model more efficiently than the simple method using ``AND'' search of queries.

  • 宮崎 遼, Paul Hannibal, 小南 貴雅, Ricardo Rosales Martinez, Borwonpob Sumeth ...
    2022 年 40 巻 2 号 p. 170-173
    発行日: 2022年
    公開日: 2022/03/17
    ジャーナル フリー

    In this study, a multi-rotor UAV equipped with an add-on planar translational driving system (ATD) was used to realize high-pressure cleaning work at high altitude. The ATD module consists of three ducted fans, which can generate force in all directions on the horizontal plane as a combined force, and can move in translation while maintaining the horizontal posture of the multirotor. In order to realize high-pressure washing work, it is required to generate appropriate reaction force to support the high-pressure washer nozzle, which sprays high-pressure water, and to accurately spray water toward the target position. In this study, we propose a method to precisely spray water by installing an ATD module in the aircraft and using its driving force to suppress the reaction force and move the aircraft while keeping its posture horizontal. In this study, we developed an aircraft that can perform high-pressure cleaning work, and verified the effectiveness of the proposed method through the experiment.

  • 福本 晃汰, 小林 泰介, 杉本 謙二
    2022 年 40 巻 2 号 p. 174-177
    発行日: 2022年
    公開日: 2022/03/17
    ジャーナル フリー

    Model Predictive Control (MPC) is one of the effective control methods for complex systems such as automatic driving and robotics. As one of the MPC solvers, the cross-entropy method (CEM) is well known as the most flexible and general method. Although CEM can be applied to most systems, it requires a sufficient (theoretically infinite) number of samples and updates for convergence, resulting in extremely high computational cost. Therefore, we focus on the asymmetry of the Kullback-Leibler divergence used in the minimization problem of CEM, and propose a new algorithm for CEM by redefining its minimization problem, so-called risk aversion CEM (RA-CEM). RA-CEM allows the function that can be regarded as a weight for the sampled trajectory to take negative values, so that even with a small iteration, the algorithm actively avoids trajectories with poor performance and prioritizes convergence to trajectories with good performance. In a highway driving simulation, RA-CEM improved the success rate from the standard CEM.

feedback
Top