Nieuwenhuisen, Matthias: Planning and Navigation in Dynamic Environments for Mobile Robots and Micro Aerial Vehicles. - Bonn, 2019. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5n-56018
@phdthesis{handle:20.500.11811/8084,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5n-56018,
author = {{Matthias Nieuwenhuisen}},
title = {Planning and Navigation in Dynamic Environments for Mobile Robots and Micro Aerial Vehicles},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2019,
month = oct,

note = {Reliable and robust navigation planning and obstacle avoidance is key for the autonomous operation of mobile robots. In contrast to stationary industrial robots that often operate in controlled spaces, planning for mobile robots has to take changing environments and uncertainties into account during plan execution. In this thesis, planning and obstacle avoidance techniques are proposed for a variety of ground and aerial robots. Common to most of the presented approaches is the exploitation of the nature of the underlying problem to achieve short planning times by using multiresolution or hierarchical approaches. Short planning times allow for continuous and fast replanning to take the uncertainty in the environment and robot motion execution into account. The proposed approaches are evaluated in simulation and real-world experiments.
The first part of this thesis addresses planning for mobile ground robots. One contribution is an approach to grasp and object removal planning to pick objects from a transport box with a mobile manipulation robot. In a multistage process, infeasible grasps are pruned in offline and online processing steps. Collision-free endeffector trajectories are planned to the remaining grasps until a valid removal trajectory can be found. An object-centric local multiresolution representation accelerates trajectory planning. The mobile manipulation components are evaluated in an integrated mobile bin-picking system.
Local multiresolution planning is employed for path planning for humanoid soccer robots as well. The used Nao robot is equipped with only relatively low computing power. A resource-efficient path planner including the anticipated movements of opponents on the field is developed as part of this thesis. In soccer games an important subproblem is to reach a position behind the ball to dribble or kick it towards the goal. By the assumption that the opponents have the same intention, an explicit representation of their movements is possible. This leads to paths that facilitate the robot to reach its target position with a higher probability without being disturbed by the other robot. The evaluation for the planner is performed in a physics-based soccer simulation.
The second part of this thesis covers planning and obstacle avoidance for micro aerial vehicles (MAVs), in particular multirotors. To reduce the planning complexity, the planning problem is split into a hierarchy of planners running on different levels of abstraction, i.e., from abstract to detailed environment descriptions and from coarse to fine plans. A complete planning hierarchy for MAVs is presented, from mission planners for multiple application domains to low-level obstacle avoidance. Missions planned on the top layer are executed by means of coupled allocentric and egocentric path planning. Planning is accelerated by global and local multiresolution representations. The planners can take multiple objectives into account in addition to obstacle costs and path length, e.g., sensor constraints.
The path planners are supplemented by trajectory optimization to achieve dynamically feasible trajectories that can be executed by the underlying controller at higher velocities. With the initialization techniques presented in this thesis, the convergence of the optimization problem is expedited. Furthermore, frequent reoptimization of the initial trajectory allows for the reaction to changes in the environment without planning and optimizing a complete new trajectory.
Fast, reactive obstacle avoidance based on artificial potential fields acts as a safety layer in the presented hierarchy. The obstacle avoidance layer employs egocentric sensor data and can operate at the data acquisition frequency of up to 40 Hz. It can slow-down and stop the MAVs in front of obstacles as well as avoid approaching dynamic obstacles.
We evaluate our planning and navigation hierarchy in simulation and with a variety of MAVs in real-world applications, especially outdoor mapping missions, chimney and building inspection, and automated stocktaking.},

url = {https://hdl.handle.net/20.500.11811/8084}
}

The following license files are associated with this item:

InCopyright