Vision Based Self-Localization Of Humanoid Robot

Self-localization is the the key problem for achieving an autonomous mobile robot. With the advancing of mobile robot self-localization research, there are new trends in the aspects of map type, sensing action planning and robot type. This thesis implemented several key algorithms of vision system based on RoboCup humanoid robot platform, and focused on the subject about the self-localization and active localization of humanoid robot. A particle filter based self-localization algorithm is implemented and a POMCP based active localization method is proposed to improve the localization efficiency.The main aspects of this thesis are as below:1. Improved image segmentation algorithm and object localization approach are proposed based on the ZJUDancer humanoid soccer robot platform.2. The self-localization problem of humanoid robot is modeled with the consideration of humanoid movement. A particle filter based self-localization algorithm is implemented, and experiment is conducted to validate the algorithm.3. The active localization problem is modeled under the POMDP(Partial Observable Markov Decision Proces) framework, and a POMCP(Partial Observable Monte Carlo Planning) based active localization algorithm is proposed. simulation experiments are conducted to validate the proposed approach and show the results of different reward functions.4. With the semantic map for humanoid soccer robot system, the POMCP-based active localization algorithm is applied for sensing action planning. Experiments are conducted to validate improvement of localization efficiency.