摘要
本論文針對人形機器人之機械手臂，提出一個靜態避障之路徑規劃法。本論文先用RGB-D攝影機擷取機器人工作區的深度影像，再將其轉換成點雲資訊，並對工作區的障礙物進行物件分割，進而得到各個障礙物的位置資訊。然後，本論文使用快速搜索隨機樹(Rapidly-exploring Random Trees, RRT) 作為基礎的路徑搜索，雖然RRT在高維空間具有極高的搜索效率來快速地找到路徑，但其易產生冗餘且崎嶇的路徑。因此本論文使用直線濾波器移除不必要的路徑點來去除迂迴路徑。為了避免馬達因為路徑震盪而造成損耗，本論文使用非均勻B樣(Non-Uniform Rational B-Spline, NURBS)曲線平滑所找到的路徑，並根據障礙物的遠近改變NURBS內的權重值。最後再藉由本論文所推導的運動學來控制機械手臂的末端點，使其跟隨規劃出來的路徑移動。從模擬與實驗結果中可得知，本論文所提出之方法確實可以讓機械手臂之末端點沿著所規劃的路徑移動，並且在不碰撞到障礙物的情況下移動到終點。
並列摘要
In this thesis, a path planning method of static obstacle avoidance is proposed for a humanoid robot arm. The depth image of the work area of robot is first captured by RGB-D camera and then converted into point cloud information. The obstacles in the work area are segmented into objects and the position of each obstacle is obtained. Then, Rapidly-exploring Random Trees (RRT) is used as a basic path search algorithm. Although RRT can find a path quickly in a high-dimensional space with high search efficiency, it tends to produce redundant and rugged path. Therefore, Line of Sight Filter is used to remove unnecessary path points to remove detour path. In order to avoid the loss of the motor due to the path oscillation, Non-Uniform Rational B-Spline (NURBS) curve is used to smooth the found path and the weight values in NURBS are changed according to the obstacle's distance. Finally, the end-effector of the robot arm is controlled based on the derived kinematics so that it can move along the planned path. From the simulation and experimental results, we can see that the proposed method can really let the end-effector of the robot arm move along the planned path and move to the end point without colliding with obstacles.