Stability and self-organization of cooperative motions in multi-fingered robot hands

多机器人协作运动的智能控制需要复杂的臂间和指间协调。虽然生物提供了概念验证，但我们对如何在控制机器人机械装置中实现概念验证的理解并不完整。主要的研究目标是理解和实现产生稳定的协调运动模式的机制，以及发展协作运动的自组织原理。在第一部分中，我们研究了构型和力相关稳定性。为了保证系统的稳定性，人们需要了解如何利用手臂的顺应性和肌肉骨骼系统的内力调节来建立稳定的运动模式。通过对运动冗余度和力冗余度的分解，得到了内力分布稳定的充要条件，并建立了可用于阻抗控制方案设计的稳定性判据。在第二部分，我们讨论了分散协调策略和合作控制。为了避免集中控制的复杂性，我们开发了一种基于遗传的机器学习技术，其强化信号建立在稳定性标准之上。初步实验证明了该概念的可行性。最后，我们研究了一个力的分布问题，找到了一个自然的、类人的力产生的最优性准则。

Intelligent control of multiple-robots in cooperative motions requires sophisticated inter-arm andinter-finger coordination. While living creatures offer proof-of-concept, our understanding of how it can be realized in controlling robotic mechanisms is not complete. The main research goals were directed to understanding and realization of mechanisms for generating stable coordinated motion patterns, and also to developing principles of self-organization of cooperative motions. In the first part we studied configuration and force-dependent stability. To guarantee the system stability, one needs to understand how the arm compliance and regulation of the internal forces in musculo-skeletal system can be used in building stable motion patterns. Decomposing the motion and force redundancy, we have found some necessary and sufficient stability conditions for the internal force distributions and established criteria of stabilizability that can be used in designing impedance control schemes. In the second part we were dealing with decentralized coordination strategy and cooperative control. To avoid the complexity of the centralized control, we have developed a genetic-based machine learning technique, with reinforcement signals built upon the stability criteria. Initial experiments show the feasibility of this concept. Finally, we have studied a force distribution problem and found an optimality criterion for natural, human-like force generation.