Please use this identifier to cite or link to this item:
|Title:||Performance evaluation of local planner algorithms for assistive wheelchairs in dense crowds||Authors:||Chua, Bok Leong||Keywords:||Engineering::Mechanical engineering::Robots||Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Chua, B. L. (2022). Performance evaluation of local planner algorithms for assistive wheelchairs in dense crowds. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/159131||Abstract:||Assistive wheelchairs are an invaluable tool in assisting people with movement difficulties in gaining back some semblance of mobility in their lives. However, one problem encountered by wheelchair users is the difficulty they face in navigating through dense crowds. Autonomous assistive wheelchairs aim to solve this problem by offloading the navigation decisions from the user to a computer, however current state-of-the-art local planner algorithms like the Dynamic Window Approach (DWA) or Timed-Elastic Band (TEB) algorithm often meet with the freezing robot problem when attempting to navigate in a dense crowd. In recent years, new algorithms based on deep reinforcement learning have shown promise in providing safe and efficient navigation through crowded spaces. A notable example is the Collision Avoidance with Deep Reinforcement Learning (CADRL) algorithm as its authors have demonstrated the algorithm in a real-world setting. What is missing is a way to compare the performance of these algorithms against one another to determine if these new algorithms have the potential to replace the current state-of-the-art. Thus, this report presents an evaluation system that quantifies the navigation performance of different local planner algorithms in a dense crowd. This evaluation system consists of four modules, namely a dense crowd simulator, a data collection module, a data processing module, and a data visualization module. The system was used to evaluate the performance of 3 algorithms (DWA, TEB, and CADRL) as well as 5 human participants. This evaluation system revealed that CADRL refused to navigate through the dense crowd, instead opting to navigate around it. Additionally, comparing the current algorithms, TEB performed 2 better than DWA in this particular dense crowd scenario. Lastly, the human participants performed similar or better in general than all three algorithms.||URI:||https://hdl.handle.net/10356/159131||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||MAE Student Reports (FYP/IA/PA/PI)|
Updated on Dec 9, 2022
Updated on Dec 9, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.