Wada Yasuhiro, Kawato Mitsuo
Nagaoka University of Technology, 1603-1 Kamitomioka, Nagaoka-shi, Niigata 940-2188, Japan.
Neural Netw. 2004 Apr;17(3):353-64. doi: 10.1016/j.neunet.2003.11.009.
In our previous research, we proposed a method for the reproduction of complex movement trajectories and robot arm control that could mimic fast, skilled human movements. That method is based on bi-directional theory and uses a representation of a set of via-points as boundary conditions or control variables to perform robot arm trajectory control. The via-points are extracted from human movement data and the resultant via-point representation is able to regenerate handwritten characters, control a Kendama toy, and perform a tennis serve. The via-point information contains both spatial and temporal information, that is, the position on the trajectory and the time of passing through the via-point position, respectively. Trajectory generation is performed using the trajectory formation model based on the optimal criterion, namely, the smoothness criterion, for which the boundary conditions are both the position and the timing of the via-point information. However, generating a smooth trajectory at different movement speeds is quite difficult if the time of passing through the via-point position is unknown or different from the extracted via-point time. In this paper, we therefore propose a new algorithm which can determine temporal via-point information. Our proposed algorithm can generate roughly the same trajectory as the measured human trajectory from only the spatial information of via-point locations. The optimality and the convergence of the new algorithm are investigated theoretically, and the trajectory generated by the algorithm is shown in numerical experiments. It is shown that starting from arbitrary temporal information the proposed algorithm can produce an appropriate trajectory.
在我们之前的研究中，我们提出了一种用于再现复杂运动轨迹和机器人手臂控制的方法，该方法可以模仿快速、熟练的人类动作。该方法基于双向理论，并使用一组过点作为边界条件或控制变量来执行机器人手臂轨迹控制。过点是从人类运动数据中提取的，所得的过点表示能够再生手写字符、控制剑玉玩具并执行网球发球动作。过点信息包含空间和时间信息，即轨迹上的位置以及通过过点位置的时间。轨迹生成是使用基于最优准则（即平滑准则）的轨迹形成模型来执行的，该模型的边界条件是过点信息的位置和时间。然而，如果通过过点位置的时间未知或与提取的过点时间不同，那么在不同运动速度下生成平滑轨迹将非常困难。因此，在本文中，我们提出了一种可以确定时间过点信息的新算法。我们提出的算法仅从过点位置的空间信息就能生成与测量的人类轨迹大致相同的轨迹。从理论上研究了新算法的最优性和收敛性，并在数值实验中展示了该算法生成的轨迹。结果表明，从任意时间信息开始，所提出的算法都能产生合适的轨迹。
