跳转到内容

避障 (机器人学):修订间差异

维基百科,自由的百科全书
删除的内容 添加的内容
无编辑摘要
标签添加文件
第1行: 第1行:
{{orphan|time=2017-10-20T17:12:01+00:00}}
{{orphan|time=2017-10-20T17:12:01+00:00}}
{{Translating|||time=2024-02-13T16:03:21+00:00}}
'''避障'''(obstacle avoidance)是[[机器人学]]中,在控制任務中附加不能接觸或是碰撞特定位置({{le|障礙|Obstacle}})等的限制條件。[[無人航空載具]]中的障礙避免是研究的熱區。在無人航空載具的都市應用中,障礙避免越來越重要,尤其是在軍事(城市戰)的應用中。一般認為障礙避免和[[运动规划]]不同,障礙避免一般會是反應性的控制律,而运动规划是{{le|預運算|Precomputatio}}一個沒有障礙的路徑,再由控制器引導機器人在預運算的路徑上。
'''避障'''(obstacle avoidance)是[[机器人学]]中自動導航以及控制系統的重要概念。是[[机器人]]或自主系統可以偵測路徑上不能接觸、不能碰撞的特定位置({{le|障礙|Obstacle}}),並且繞過障礙以達到事先定義的目的地。此技術在工業自動化、自駕車、無人飛機其至太空船上都相當的重要。避障讓機器人可以在動態以及複雜的環境下安全及有效率的運作,減少碰撞以及破壞的風險。

机器人及自主系統要成功的在有障礙物的情形下導航,必須要可以偵測這些障礙物,這大部份會用[[感測器]]來進行,感測器送出的資料讓機器人可以處理周圍環境資訊,決定要怎麼走才能避開障礙物,並且利用致動器(或是其他機器人和環境互動的工具)執行這些決定<ref name=":0">{{cite book | vauthors = Wang J, Herath D | chapter = What Makes Robots? Sensors, Actuators, and Algorithms |date=2022 | title = Foundations of Robotics: A Multidisciplinary Approach with Python and ROS |pages=177–203 | veditors = Herath D, St-Onge D |place=Singapore |publisher=Springer Nature |language=en |doi=10.1007/978-981-19-1983-1_7 |isbn=978-981-19-1983-1 }}</ref>。

== 作法 ==
机器人或自主系統要實時進行避障決策的作法有幾種。其中包括有以感測器以基礎的作法、[[运动规划]]演算法,以及[[机器学习]]技術。
=== 感測器以基礎 ===
[[File:01Cruise-E-Ultrasonic_Intelligent_Obstacle_Avoidance.gif|thumb|利用感測器進行的避障避障]]
避障技術最常見的作法是配合許多的感測器,例如[[超音波]]、[[光学雷达]]、[[雷達]]、[[声呐]]以及[[攝影機]]。這些感測器讓自主系統可以進行三步驟的流程:感測、思考、行動。系統會取得不同物件的距離輸入,作法是將机器人周遭的資料提供給机器人,讓機器人偵測障礙物並計算距離。接著機器人就可以在維持計算路徑的前題下,小幅調整軌跡以繞過障礙物,大部份的避障應用程式可以實時完成這些步驟,並且以務實並且有效率的方式進行<ref name=":0" /><ref>{{cite conference | vauthors = Discant A, Rogozan A, Rusu C, Bensrhair A | title = Sensors for obstacle detection-a survey. | conference = 2007 30th International Spring Seminar on Electronics Technology (ISSE) | date = May 2007 | pages = 100–105 | publisher = IEEE | doi = 10.1109/ISSE.2007.4432828 }}</ref>。

上述方式在大部份的情形下都有很好的效率。不過有些更先進的技術可以使用,適用於需要有效率的到達終點的應用。
[[File:Weighted_A_star_with_eps_5.gif|thumb|A*路徑規劃演算法的例子]]

=== 路徑規劃演算法 ===
{{transh}}
Path Planning Algorithms are critical for optimally calculating and routing collision-free paths. These algorithms take into account the robot's position, destination, and the locations of obstacles in the environment. They take and store this information to map out an area, and then use that map to calculate the fastest possible route to a specific destination. Such algorithms are commonly used in routing mazes and autonomous vehicles. Popular path-planning algorithms include [[A* search algorithm|A*]] (A-star), [[Dijkstra's algorithm]], and [[Rapidly exploring random tree|Rapidly-exploring Random Trees]] (RRT). These algorithms help the robot find the quickest path to reach its goal while avoiding collisions, all in real time.<ref>{{cite journal | vauthors = Véras LG, Medeiros FL, Guimaráes LN | title = Systematic literature review of sampling process in rapidly-exploring random trees. | journal = IEEE Access | date = March 2019 | volume = 7 | pages = 50933–50953 | doi = 10.1109/ACCESS.2019.2908100 | bibcode = 2019IEEEA...750933V | s2cid = 133481997 | doi-access = free }}</ref>
{{transf}}

=== 机器学习技術 ===
{{transh}}
With the use of machine learning, the range of possibilities for obstacle avoidance becomes far greater. With artificial [[Artificial intelligence|intelligence]] (AI), an autonomous machine can figure out a path to get to its destination, but can also learn to adapt to a rapidly changing environment at the same time. It can do this by being put through many testing stages of exposure to obstacles and environmental changes. By giving an AI a task and reward for doing a task correctly, over time, it can learn to do this task efficiently and effectively. This allows the machine to understand what its obstacles are and to come up with an efficient path around them. It also gives the machine the ability to learn how to deal with specific cases, which can include dealing with water, hills, high winds or temperatures, etc. This use of AI allows the autonomous machine to react accordingly to a plethora of situations that could be expected or unexpected. This form of obstacle avoidance is especially good in autonomous vehicles as it removes possible human errors that can occur. <ref>{{Cite journal | vauthors = Bachute MR, Subhedar JM |date=December 2021 |title=Autonomous Driving Architectures: Insights of Machine Learning and Deep Learning Algorithms |journal=Machine Learning with Applications |language=en |volume=6 |pages=100164 |doi=10.1016/j.mlwa.2021.100164|s2cid=240502983 |doi-access=free }}</ref>
{{transf}}


==相關條目==
==相關條目==
第7行: 第29行:
* {{le|機器人控制|Robot control}}
* {{le|機器人控制|Robot control}}


==參考資料==
{{reflist}}
== 外部連結 ==
== 外部連結 ==
* [http://www.asirobots.com/research/ Forecast 3D Laser System] {{Wayback|url=http://www.asirobots.com/research/ |date=20210515194718 }}: a LIDAR based obstacle detection and avoidance sensor. Forecast generates a 3D [[point cloud]] or cost map output that can be used for robotic command and control software, terrain mapping, and other applications.
* [http://www.asirobots.com/research/ Forecast 3D Laser System] {{Wayback|url=http://www.asirobots.com/research/ |date=20210515194718 }}: a LIDAR based obstacle detection and avoidance sensor. Forecast generates a 3D point cloud or cost map output that can be used for robotic command and control software, terrain mapping, and other applications.


== 延伸閱讀 ==
== 延伸閱讀 ==
第15行: 第39行:
[[Category:控制理论]]
[[Category:控制理论]]
[[Category:机器人导航]]
[[Category:机器人导航]]


{{技術小作品}}

2024年2月13日 (二) 16:03的版本

避障(obstacle avoidance)是机器人学中自動導航以及控制系統的重要概念。是机器人或自主系統可以偵測路徑上不能接觸、不能碰撞的特定位置(障礙英语Obstacle),並且繞過障礙以達到事先定義的目的地。此技術在工業自動化、自駕車、無人飛機其至太空船上都相當的重要。避障讓機器人可以在動態以及複雜的環境下安全及有效率的運作,減少碰撞以及破壞的風險。

机器人及自主系統要成功的在有障礙物的情形下導航,必須要可以偵測這些障礙物,這大部份會用感測器來進行,感測器送出的資料讓機器人可以處理周圍環境資訊,決定要怎麼走才能避開障礙物,並且利用致動器(或是其他機器人和環境互動的工具)執行這些決定[1]

作法

机器人或自主系統要實時進行避障決策的作法有幾種。其中包括有以感測器以基礎的作法、运动规划演算法,以及机器学习技術。

感測器以基礎

利用感測器進行的避障避障

避障技術最常見的作法是配合許多的感測器,例如超音波光学雷达雷達声呐以及攝影機。這些感測器讓自主系統可以進行三步驟的流程:感測、思考、行動。系統會取得不同物件的距離輸入,作法是將机器人周遭的資料提供給机器人,讓機器人偵測障礙物並計算距離。接著機器人就可以在維持計算路徑的前題下,小幅調整軌跡以繞過障礙物,大部份的避障應用程式可以實時完成這些步驟,並且以務實並且有效率的方式進行[1][2]

上述方式在大部份的情形下都有很好的效率。不過有些更先進的技術可以使用,適用於需要有效率的到達終點的應用。

A*路徑規劃演算法的例子

路徑規劃演算法

机器学习技術

相關條目

參考資料

  1. ^ 1.0 1.1 Wang J, Herath D. What Makes Robots? Sensors, Actuators, and Algorithms. Herath D, St-Onge D (编). Foundations of Robotics: A Multidisciplinary Approach with Python and ROS. Singapore: Springer Nature. 2022: 177–203. ISBN 978-981-19-1983-1. doi:10.1007/978-981-19-1983-1_7 (英语). 
  2. ^ Discant A, Rogozan A, Rusu C, Bensrhair A. Sensors for obstacle detection-a survey.. 2007 30th International Spring Seminar on Electronics Technology (ISSE). IEEE: 100–105. May 2007. doi:10.1109/ISSE.2007.4432828. 
  3. ^ Véras LG, Medeiros FL, Guimaráes LN. Systematic literature review of sampling process in rapidly-exploring random trees.. IEEE Access. March 2019, 7: 50933–50953. Bibcode:2019IEEEA...750933V. S2CID 133481997. doi:10.1109/ACCESS.2019.2908100可免费查阅. 
  4. ^ Bachute MR, Subhedar JM. Autonomous Driving Architectures: Insights of Machine Learning and Deep Learning Algorithms. Machine Learning with Applications. December 2021, 6: 100164. S2CID 240502983. doi:10.1016/j.mlwa.2021.100164可免费查阅 (英语). 

外部連結

延伸閱讀

  • BECKER, M. ; DANTAS, Carolina Meirelles ; MACEDO, Weber Perdigão, "Obstacle Avoidance Procedure for Mobile Robots". In: Paulo Eigi Miyagi; Oswaldo Horikawa; Emilia Villani. (Org.). ABCM Symposium Series in Mechatronics, Volume 2. 1 ed. São Paulo - SP: ABCM, 2006, v. 2, p. 250-257. ISBN 978-85-85769-26-0