Sampling-based path planning algorithm have struggle dealing with environment with narrow regions. Because sampling is done for the entire configuration space rather than recognizing difficult region and approach selectively. Therefore many configurations are formed around the narrow regions not getting samples inside of it. Although many method has developed to solve the narrow passage problem, the computational cost was high to figure out narrow region precisely. Also, there was a risk for misreconizing the region types and failed to adjust in narrow passage with small free space. Thus, to overcome these drawbacks, the proposed method classifies regions which have a potential to be surface or narrow region in a probabilistic manner. Based on RRT-Connect, the roadmap is formed with two trees expanded by visiting the classified potential regions. Therefore, collecting and using the samples inside of narrow regions is possible without retraction or additional tests. As a result, the proposed method was verified on 2D and 3D maps with narrow passage compared to the original RRT-Connect. This method showed good performance especially on maps with many direction changes or very narrow regions. The performance was improved 54.8% for 2D maps and 64.3% for 3D maps in average.
|Udgiver||Department of Interdisciplinary Engineering Systems, Graduate School of Hanyang University|
|Status||Udgivet - feb. 2020|
Bibliografisk noteThesis Supervisor: Professor Kyoo Sik Shin
Song, H. (2020, feb). Improved path planning algorithm based on RRT-Connect guided by probabilistically estimated collision region. Department of Interdisciplinary Engineering Systems, Graduate School of Hanyang University.