A Study Of Footstep Planning For A Humanoid Robot Based On Visual Identification

Research on humanoid robot is a challenging area in robotic research. With the ad-vancement of biped-walking pattern planning and balance maintenance, intelligent percep-tion and control of humanoid robots become hot research topics. In this dissertation deep insights were given into the research on visual identification, footstep planning and path optimization technology of humanoid robot, including:1. An image-segmentation approach and a geometrical model were presented for visual identification and localization. The color region can be segmented by tracking the boundary, then the region's features are abstracted from the boundary geometrically. The localization model combines the pin-hole camera model and the physical link model of robots, and can localize the object precisely. The method and model were well applied in humanoid robot soccer match.2. A dynamic threshold raceway identification approach was used to aim at the visual identification of outdoor environment. The threshold is obtained from the histogram of RGB space to filter the irrelevant color. Then the raceway is recognized using edge detection and line fitting. The application of walking in the playground demonstrated that the approach is robust to varying light.3. A new algorithm named Heuristic Bi-directional Discrete Rapidly-explore Random Tree (HBD-RRT) was presented for navigation problem in obstacle-cluttered envi-ronment. According to discrete stance model, two randomized-sampling-based plan-ning trees from start state and target state are constructed respectively, where the path is generated as a set of stances by bi-directional search. Then a series of off-line gaits are used to transform the path to motion trajectory. The simulation result in the plat-form of ZJU-Dancer humanoid robot demonstrated that HBD-RRT for biped robots is less-time-consuming. 4. To plan a smooth path and walk in dynamic-obstacle environment, the elastic band algorithm was extended to optimize the path. A virtual force is created for centroid of robot to adjust the smoothness of the path. The update of centroid is subject to the discrete stance model and renews the stance to make a smooth path. The appli-cation of purchasing football and tracking raceway in playground demonstrated the algorithm can smooth the path and avoid the obstacle effectively.