节律-技巧混合驱动的机器人行走机理研究及实验验证

项目名称： 节律-技巧混合驱动的机器人行走机理研究及实验验证

项目编号： No.60875057

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 生物科学

项目作者： 陈启军

作者单位： 同济大学

项目金额： 27万元

中文摘要： 机器人行走的环境适应性问题是机器人现场应用的瓶颈，本项目基于CPG机理来解决机器人行走控制及环境适应性问题，取得了满意的结果：（1）提出了节律-技巧混合驱动策略，节律信号由CPG来实现，通过技巧学习来耦合感知信息，达到了实时调控关节控制信号的目的；（2）提出了基于进化算法的交互式启发式算法，充分集成了计算智能和人类智能，取得了满意的优化结果；（3）在关节空间，提出了运动映射的控制方法，降低了传统关节空间控制方法的复杂性；（4）针对四足机器人，提出了将CPG网络输出的一维信号映射为三维工作空间轨迹的映射方法，提高了行走稳定性；（5）针对两足机器人，提出了重心轨迹规划和工作空间轨迹调制方法。利用CPG的极限环和携带特性，吸收传感反射信息，获得具有环境适应性的工作空间轨迹；（6）开发出一套机器人技巧学习和行为融合的调试平台，提高算法测试速度；（7）自主研发了一套仿人机器人的全向行走在线调试软件，提高了算法测试的灵活性。本项目在国际上首次成功的针对实体机器人完成了生物诱导适应性行走控制实验，充分显示了研究思路的有效性。这对突破现场机器人真实应用的瓶颈，促进服务机器人的开发和应用都具有重要意义。

中文关键词： 机器人;技巧学习;混合驱动;行走控制;中枢模式发生器

英文摘要： The environmental adaptability problem of locomotion control is the bottleneck of application of legged robots in real world. This project develops locomotion control methods based on central pattern generator (CPG) mechanism to solve this problem, and the experiments achieved satisfactory results: (1) By combining rhythmic movements and skills learning, a hybrid-driven control strategy is presented, where rhythmic movements are realized through CPG and skills are gain from a learning process which are used to adjust the control signals for the joints in-real time; (2) The parameters of the system are evolved by interactive heuristic algorithm which based on multi-objective evolutionary algorithm, and the evolution achieved satisfactory results; (3) A motion mapping method is presented in joint space of the robot to reduce the complexity of the traditional joint-space control method; (4) For quadruped, a novel control method is proposed which using CPGs to generate workspace trajectories. A mapping strategy is proposed to map the output signals of CPGs to three-dimensional workspace trajectories online to improve the adaptability of the locomotion; (5) For biped, a center of gravity trajectories generated method and a workspace trajectories modulated method are proposed. Moreover, by using the limit cycle and entrainment properties of the CPGs to realize the adaptive trajectories online through sensory feedback adjustment; (6) A test-bed for simulation and debugging of the skill learning and behavior fusion based is developed, which accelerated the research process; (7) A test-bed for simulation and debugging of the omnidirectional walking of humanoid is developed, which improves the flexibility of the test process. This project completes the first successfull experiment of adaptive locomotion control of real legged robots inspired by CPG-based biological concept, which validates the effectiveness of our research. The investigation of this project is very important for breaking the bottleneck of application of field robots, and it is very important both in exploitation and application for service robots.

英文关键词： robot; skill learning; hybrid-driven; locomotion control; central pattern generator(CPG)