AN IMPROVED VISION-BASED SLAM APPROACH INSPIRED FROM ANIMAL SPATIAL COGNITION

摘要

Simultaneous Localization and Mapping (SLAM) is not only an important task but also a challenge in the robotic field. Recently, vision-based SLAM approach has become the research hot spot. There are some disadvantages of the general vision-based SLAM algorithm such as the higher computational requirements and the lower accuracy than the probability-based SLAM method. To deal with these problems, an improved vision-based SLAM approach is proposed based on the RatSLAM method, which is inspired from animal spatial cognition. In the proposed approach, a local search strategy is used to improve the real-time performance of the general RatSLAM method. In order to get a more accurate map, a prediction strategy is added to the general RatSLAM method, then the robot can predict the next possible view or position, which can reduce the effects of the noise and accumulated error on the SLAM method. In addition, a concept of the short-term memory map is introduced into the proposed SLAM method, to solve the kidnap recovery problems. Finally, various real robot SLAM experiments are conducted, and the results show that the proposed vision-based SLAM approach is more efficient than the general RatSLAM method.