臺灣大學: 電機工程學研究所羅仁權張宏豪Chang, Hong-HaoHong-HaoChang2013-03-272018-07-062013-03-272018-07-062012http://ntur.lib.ntu.edu.tw//handle/246246/253887本研究重點在使雙足機器人產生一個較接近人類平常行走的步態，外國學者曾有實際測量人類在行走時ZMP(零力矩點)的變化，我們發現和大部分研究所規劃的ZMP有相當大的差異性，因此，我們試著實作出一組ZMP規劃如同人類一般行走的變化，使得雙足機器人行走的行為更接近於人類。 本研究所使用的雙足機器人數學模型為三質點模型，相較於一般研究所使用線性倒單擺模型和重力補償倒單擺模型，更能夠充分的代表雙足機器人的物理結構。 在步態發展上使用基於ZMP理論的步態軌跡產生器，並將發展的步態區分成三段流程 : 分別是下蹲，升起，和倒單擺模式。在步態的特色上具有以下幾個特色 : (1)身體軌跡近似於等速移動。 (2) ZMP規劃始終向前移動，沒有卡在某個位置的情形。 (3) 身體高度盡量地拉高，並在走路的過程並非維持相同高度。 綜合以上三個特色，我們可以發展出較自然且好看的行走步態。同時，在步態參數的調整上，我們提供了一套參數調整的準則，利於控制機器人行走的步伐大小和速度。 在雙足機器人的行走實驗上，我們使用Windows XP為作業系統，配合RTX這套軟體提供高度精準且快速取樣時間，再搭配常見的編碼器/解碼器，數位類比轉換器所組成的運動控制卡，構成了一套非常利於開發的控制系統。本實驗實作了步伐長度5公分到30公分的幾組步態，從截取的實驗過程片段照片裡可以發現我們所發展的步態具有相當程度的穩定性，而在視覺效果上給予了和大部分研究所使用的步態相當不一樣的感受，是一個膝蓋較自然走路的步態。同時也利用我們的步態產生方法，實作了一般身體腰部在同一水平面的步態。由這些不同的步態所得到的實際ZMP回授的數據所示，雖然離預先規劃的ZMP尚有一段誤差，但也說明了我們發展的步態規劃方法具有一定程度的穩定性。The objective of this research is mainly in generating a biped robot walking pattern towards more closer to mimic human walking. In many research, we found that the ZMP(Zero Moment Point) planning of their walking trajectory has a big difference between that measured from human walking. Therefore, we tried to implement a biped robot walking pattern to with the ZMP planning mimic to human walking. The mathematical model of the biped robot used in our research is composed of three main points, which is very different from LIPM (Linear Inverted Pendulum Model) and GCIPM(Gravity Compensated Inverted Pendulum Model). This model represents the physical structure of biped robot better than LIPM and GCIPM. We use ZMP based trajectory generator with the three mass model and divide walking behavior into three processes(lowering down, rising up and inverted pendulum mode processes). The walking pattern we developed has the following features: (1) quasi-constant speed body movement. (2) The ZMP planning goes monotonically forward with no intermediate stop. (3) The height of body is quite high and it is at non-constant height during the walking process. Combining the three features above, we can develop a quasi natural walking pattern. In the mean time, we also provided a rule of how to set the parameters of walking trajectory generator and it makes us easier to control the stride length and walking velocity. In the biped robot walking experiment, we use Windows XP as the operating system with RTX real time kernel. In addition, with common modules such as encoders/decoders and AD/DA converter, we can easily establish a control system that is very friendly to developers. In our experiment, walking patterns with stride length of 5cm to 30cm is implemented. From the snapshots of the experiment, we find that the walking behavior has a big difference from other ZMP based trajectory generating methods and is quite stable. With the same method, we can get a common walking pattern at constant body height that is similar to the walking behavior of HRP and ASIMO. From the ZMP response measured in the experiment, we know that there are big errors from the desired ZMP reference but in the mean time the stability of the proposed walking pattern has been proved.3327042 bytesapplication/pdfen-US雙足控制bipedcontrolthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/253887/1/ntu-101-R99921006-1.pdf