國立虎尾科技大學 |

Modeling Framework for Collision Avoidance Learning in Multirotor Unmanned Aerial Variables.

紀錄類型:	書目-語言資料,印刷品 : Monograph/item
正題名/作者:	Modeling Framework for Collision Avoidance Learning in Multirotor Unmanned Aerial Variables./
作者:	Gabani, Krushang Khimjibhai.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2020,
面頁冊數:	92 p.
附註:	Source: Masters Abstracts International, Volume: 81-09.
Contained By:	Masters Abstracts International81-09.
標題:	Robotics. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27736635
ISBN:	9781658421706

Modeling Framework for Collision Avoidance Learning in Multirotor Unmanned Aerial Variables.
Gabani, Krushang Khimjibhai.

Modeling Framework for Collision Avoidance Learning in Multirotor Unmanned Aerial Variables. - Ann Arbor : ProQuest Dissertations & Theses, 2020 - 92 p.

Source: Masters Abstracts International, Volume: 81-09.

Thesis (M.Eng.)--State University of New York at Buffalo, 2020.

This item must not be sold to any third party vendors.

This work focuses on the idea of generating the learning framework for cooperative collision avoidance between two quadcopters. Two strategies for reciprocal online collision-avoiding actions (i.e., coherent maneuvers without requiring any real-time consensus) are proposed. In the first strategy, UAVs change their speed, while in the second strategy, they change their heading to avoid a collision. The avoidance actions are parameterized in terms of the time difference between detecting the collision and starting the maneuver and the amount of speed/heading change. These action parameters are used to generate intermediate way-points, subsequently translated into a minimum snap trajectory, to be executed by a PD controller. This work presents a learning-based approach to train such reciprocal maneuvers. A Neuroevolution approach, which uses evolutionary algorithms to optimize the topology and weights of neural networks simultaneously, is used as the learning method -- which operates over a set of sample approach scenarios to evaluate the fitness of each neural network candidate. Unlike most existing work (that minimize travel distance), the training process here has the capability of minimizing the required detection range and detection time. The specialized design of experiments and line search is used to identify the minimum detection range for each sample scenarios. This training process can also handle substantial challenges like uncertain UAV localization. For that, the relative pose of the other UAV, estimated by each UAV (at the point of detection), is considered to be uncertain. These capabilities have essential practical implications w.r.t. alleviating the dependency on sophisticated sensing and their reliability under various environments. Performing supervised learning based on optimization derived labels (as done in prior work) becomes computationally burdensome under these uncertainties. For an efficient training process, a classifier is used to discard actions (without simulating them) where the controller would fail. Also, a surrogate model is used to estimate the energy consumption and minimum distance between UAVs. The model obtained via neuroevolution is observed to generalize well to (i.e., successful collision avoidance over) unseen approach scenarios.

ISBN: 9781658421706Subjects--Topical Terms:

561941
Robotics.
Subjects--Index Terms:

Collision avoidance

Modeling Framework for Collision Avoidance Learning in Multirotor Unmanned Aerial Variables.
LDR:03501nam a2200361 4500 001 1037934
005 20210910100635.5
008 211029s2020 ||||||||||||||||| ||eng d
020 $a 9781658421706
035 $a (MiAaPQ)AAI27736635
035 $a AAI27736635
040 $a MiAaPQ $c MiAaPQ
100 1 $a Gabani, Krushang Khimjibhai. $0 (orcid)0000-0001-5172-1703 $3 1335241
245 1 0 $a Modeling Framework for Collision Avoidance Learning in Multirotor Unmanned Aerial Variables.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 92 p.
500 $a Source: Masters Abstracts International, Volume: 81-09.
500 $a Advisor: Chowdhury, Souma.
502 $a Thesis (M.Eng.)--State University of New York at Buffalo, 2020.
506 $a This item must not be sold to any third party vendors.
506 $a This item must not be sold to any third party vendors.
520 $a This work focuses on the idea of generating the learning framework for cooperative collision avoidance between two quadcopters. Two strategies for reciprocal online collision-avoiding actions (i.e., coherent maneuvers without requiring any real-time consensus) are proposed. In the first strategy, UAVs change their speed, while in the second strategy, they change their heading to avoid a collision. The avoidance actions are parameterized in terms of the time difference between detecting the collision and starting the maneuver and the amount of speed/heading change. These action parameters are used to generate intermediate way-points, subsequently translated into a minimum snap trajectory, to be executed by a PD controller. This work presents a learning-based approach to train such reciprocal maneuvers. A Neuroevolution approach, which uses evolutionary algorithms to optimize the topology and weights of neural networks simultaneously, is used as the learning method -- which operates over a set of sample approach scenarios to evaluate the fitness of each neural network candidate. Unlike most existing work (that minimize travel distance), the training process here has the capability of minimizing the required detection range and detection time. The specialized design of experiments and line search is used to identify the minimum detection range for each sample scenarios. This training process can also handle substantial challenges like uncertain UAV localization. For that, the relative pose of the other UAV, estimated by each UAV (at the point of detection), is considered to be uncertain. These capabilities have essential practical implications w.r.t. alleviating the dependency on sophisticated sensing and their reliability under various environments. Performing supervised learning based on optimization derived labels (as done in prior work) becomes computationally burdensome under these uncertainties. For an efficient training process, a classifier is used to discard actions (without simulating them) where the controller would fail. Also, a surrogate model is used to estimate the energy consumption and minimum distance between UAVs. The model obtained via neuroevolution is observed to generalize well to (i.e., successful collision avoidance over) unseen approach scenarios.
590 $a School code: 0656.
650 4 $a Robotics. $3 561941
650 4 $a Mechanical engineering. $3 557493
653 $a Collision avoidance
653 $a Machine learning
653 $a Quadcopter
653 $a Sensor deficiency
690 $a 0548
690 $a 0771
710 2 $a State University of New York at Buffalo. $b Mechanical and Aerospace Engineering. $3 845600
773 0 $t Masters Abstracts International $g 81-09.
790 $a 0656
791 $a M.Eng.
792 $a 2020
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27736635