Localization and motion planning of mobile robots

• Main Author: Dai, Yunxiang
• Other Authors: Wang Dan Wei
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
• Online Access: https://hdl.handle.net/10356/165275
• Institution: Nanyang Technological University

Robot navigation system is a key functional module of autonomous robots. The development of the navigation system largely depends on the working environment and tasks. Key components including localization and motion planning are usually customized according to the specific application. Therefore, how to design the application-oriented robot navigation system and integrate all the sub-modules properly is a worthy research topic. This thesis studies robot navigation in two application scenarios. The first part of the thesis introduces a navigation system intending to fulfill an autonomous route repeating and fleet maintenance problem for multiple mobile robots in the open area. An Ultra-Wideband-based robot localization method is firstly implemented to provide the pose estimation of every robot in the troop. UWB sensors are selected due to their robustness in featureless open area environments compared to other localization algorithms using Lidar or visual cameras. The localization system relies on the Extended Kalman Filter(EKF) as the backbone while is initialized by non-linear trilateration. Afterward, a record-and-repeat pipeline is designed for the single robot's route repeating tasks. The waypoints of the desired route are sampled in the record loop, and a feasible trajectory is generated accordingly in an offline algorithm and serves as the reference for the single robot route following. Finally, a leader-follower controlling mechanism is proposed to solve the multi-agent formation maintenance part of the system. The whole system as well as every sub-modules are all validated to be effective in the open area testing sites. The localization can achieve an error less than $0.15\mathrm{m}$ on average in the testing site with $50\mathrm{m}\times50\mathrm{m}$ size. Meanwhile, the offline reference trajectory generation mechanism and a low-level motion controller for both single and multiple robots are tested to work successfully under different route settings, traveling speed, and fleet configurations, which demonstrates the robustness of the motion planning sub-system. The proposed system can be applied to various cases such as unmanned inspection or autonomous transportation in seaports or industrial parks, facilitating the working efficiency and making humans free from arduous tasks. In the second part, another typical application case for autonomous driving in urban road environments is investigated. The unmanned road sweeper has been one of the first industrialized unmanned vehicle products in recent years, which usually needs to travel along the urban road curb for environmental cleaning. To handle this problem, an urban road curb following strategy is proposed to serve as part of the navigation module of the unmanned sweeper. A curb detection method is firstly applied to extract candidate curb points in a single frame, followed by a multi-frame fusion further generating representative points from the curb detection results in the single frame. The pipeline is designed in this way as a result of balancing the accuracy and real-time performance. Afterward, a novel reference path segment generation and switching mechanism are proposed to bridge the perception with the low-level controller. The well-designed mechanism also guarantees the smoothness of reference trajectories and the feasibility of robot motion under the limited perception of the road curb ahead. Finally, a path-following controller is adopted in the system to drive the robot traveling along the reference path generated in the previous step. The proposed strategy is tested, in both simulation and real-world experiments, to be adaptable to various types of urban roads, including the straight, arc-shaped, and concave-shaped roads. Both qualitative and quantitative results are presented for curb detection and the following part, respectively.