基於能量函數與倒單擺模型之雙足機器人步態軌跡規劃與控制

Abstract

在許多雙足機器人研究議題上，為了要控制行走的穩定性，髖部的軌跡通常都會規劃在同一水平面上，因此當這些機器人在行走時膝蓋總是彎曲的。然而，這一點都不像我們人類的行走方式。 為了解決這個問題，我們發展一套能使雙足機器人呈現類人式走法(膝蓋打直)且能穩定行走的步態軌跡生成演算法。首先，本研究先利用SOLIDWORKS先建立一個雙足機器人的3D模型，接著將其匯入Matlab軟體的Simmechanics的重力場環境中，接著藉由本演算法寫成的軌跡產生器來生成步態，並調整其中的兩個本演算法提出的穩定參數， 直到模擬環境中的雙足機器人能行走。最後，再將他們套用於實際雙足機器人的軌跡產生器來產生軌跡。然而，實體的雙足機器人與模擬環境中的數學模型不盡相同。因此這兩個穩定參數雖已在模擬環境中精練過，但仍需就實體雙足機器人行走的姿勢(前傾或後倒)來進行些微調整，這部分可藉由高速攝影機來擷取行走時的情況，並依照本論文提出之方法做些微調整，最後即可讓實體雙足機器人行走。 從本論文最後的模擬與實驗可證明，建立在我們提出演算法上的軌跡產生器確實能夠產生各種大小步伐和不同行走時間的穩定步態，且這些步態與世界上大部分雙足機器人(如ASIMO, HRP系列)彎著膝蓋行走的走法不同，是為直立行走的步態。 此外，從本論文的截圖與其說明中，可清楚看出各張截圖在本演算法中所代表的各個時態與含義，每張截圖皆為本理論的時間軸中的各個時間點，在這些不同的時間點中，我們可看到其對應分解動作，與人類行走方式確實有著極大的相似。

In many research of biped robotics, for controlling the walking stability of the robot, the hip trajectory is planned at the same height, so the knees of the robot would bend while walking. However, this does not make sense when humans walk. To deal with the problem, in this thesis, we develop a walking pattern generating algorithm, which would enable the biped robot walking like human (stretch knee walking). First of all, we build a 3D biped robot model by using SOLIDWORKS, and then convert it into gravitational simulation of Simmechanics of Malab software. We generate its walking pattern by the trajectory generator based on the algorithm of this research, and tune its two parameters of stability until the robot in simulation can walk. Finally, we apply the two parameters to the trajectory generator of the actual biped robot to generate the stable pattern. However, there exist modeling errors between the actual biped robot and the model in simulation. Therefore, although we have refined the two parameters in simulation, they still have to be adjusted slightly according to actual walking postures (fall down forward or backward). For this reason, we can use high speed camera to get the actual walking information, and tune the parameters slightly according to the methods in the thesis. Finally, we would enable the robot to walk. From our simulation and experiment, the trajectory generator based on our algorithm indeed can generate several stable patterns with different step length and step time, and these knee-stretching patterns are different from the most biped robots in the world (such as ASIMO and HRP series), which walk with knee-bending patterns. Furthermore, from the snapshots and their illustrations, we can understand the process and meaning in each snapshot. Each snapshot represents a moment of the time axis in our algorithm, and we can see there are many similar points with the human walking behavior from the corresponding separated behavior in each snapshot.