人形机器人是机器人研究领域里一个重要分支。人形机器人的技术发展方向是要在运动和智能方面的行为尽量与人类相似，其目标功能是要能与人类工作在同一环境里，与人类协作和提供服务，甚至能完成一些人类无法完成的任务。要实现人形机器人这一目标，涉及需要研究的领域非常广泛。本论文的工作正是基于人形机器人发展方向作为基本思路展开的：从底层的数学建模到基本动作的步态合成，再到更高层的人形机器人全局运动规划；从完成简单的运动到具有智能决策的能力；从简单直线行走的步态合成到复杂的转弯步态合成；对人形机器人的运动要求先保证基本的稳定性再到实现高性能的运动。本论文所完成的工作和主要贡献概括为以下几个方面： 1．给出了基于改进型倒立摆模型的直线行走和转弯的步态规划。通过约束倒立摆的运动保证了机器人运动的稳定性和连续性，在直线行走中通过约束机器人摆动脚在脚跟着地时踝关节的转动速度来减小机器人与地面的碰撞。同时，利用本文所提出的人形机器人逆运动学计算方法，机器人的步态合成能生成双腿12个自由度的精确参考轨迹，这些参考轨迹用来控制实际机器人运动。 2．提出了一种基于遗传算法和神经网络的步态优化方法来提高机器人运动性能，与目前其它基于遗传算法的步态优化相比，本文提出的优化算法同时考虑了机器人在横向平面和径向平面的运动对机器人运动性能的影响，并且优化目标也同时考虑了能量消耗最小和稳定域最大。用遗传算法优化了一组步态集后，又用一个神经网络将优化后的步态信息存贮和泛化，为更高层的决策做准备。 3．除了给出基本动作的步态合成和优化外，本文还利用这些基本动作作为人形机器人的局部运动实现人形机器人的全局运动，同时还阐述了人形机器人全局运动和局部运动的关系，通过切换动作保证了人形机器人在不同基本动作之间的平滑切换。人形机器人全局运动规划的输出是一条无碰撞的全局路径和实现全局路径时双腿12个自由度的参考轨迹。
Humanoid robot has been an important research topic in the field of robotics. The technology objective of humanoid robots is to build a robot that resembles human behavior regarding motion and intelligence. The objective of its function is to coexist with, collaborate with and severe humans, or even to perform some tasks that humans cannot. To achieve these objectives, a broad research fields are involved. The work of this thesis is based on the objectives of humanoid research, from the low level（mathematical modelling and gait synthesis of basic motion）to high level （global motion planning）, from realizing simple motion to being capable of making intelligent decisions, from simple straight walking to complex turning motion. The requirements of humanoid motion are also from basic stability to high performance. The main contributions of this thesis are given as follows: 1. An improved IPM (Inverted Pendulum Model) used for gait synthesis of straight walking and turning is presented in this thesis. By restricting the motion of IPM, the stability and cotinutiy of humanoid motion is achieved. The impact effects between the robot and ground are relieved by limiting the joint velocity of ankle joint at the moment when the heel of swing leg touches ground. By using the calucation method of inverse kinematics for humanoid robot given in this thesis, the gait synthesis software is able to generate accurate reference trajectories for all 12 DOF (Degree Of Freedom) of two legs of humanoid robot. These reference trajectories are used to control real robots to move.2. A GA (Genetic Algorithm) and NN (Neural Network) based optimization method for humanoid gait synthesis is proposed in this thesis to improve humanoid motion performance. Compared with other GA based humanoid optimization methods, the humanoid motions in both the sagittal and frontal planes are considered to improve walking performance. Both minimizing energy consumption and increasing stability area are taken into account. After obtaining a set of near-optimal walking gaits from GA search, a neural network is adopted to store and generalize these walking gaits information, which is prepared for higher level decision.3. Besides the gait synthesis and gait optimization for humanoid basic motion, a new approach to realizing humanoid global motion by using these basic motions as local motion is given in this thesis. In addition, the relations between humanoid global motion and local motion are clarified. The smooth transitions between different basic motions are achieved by the introduction of transition motion. The output of humanoid global motion planning is a collision-free global path and 12 reference trajectories for 12 DOF that realizes the global path.
