UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement
Deep reinforcement learning (DRL) has significantly advanced online path planning for unmanned aerial vehicles (UAVs). Nonetheless, DRL-based methods often encounter reduced performance when dealing with unfamiliar scenarios. This decline is mainly caused by the presence of non-causal and domain-spe...
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2024-12-01
|
| Series: | Drones |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2504-446X/9/1/10 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832588617541746688 |
|---|---|
| author | Zhun Fan Zihao Xia Che Lin Gaofei Han Wenji Li Dongliang Wang Yindong Chen Zhifeng Hao Ruichu Cai Jiafan Zhuang |
| author_facet | Zhun Fan Zihao Xia Che Lin Gaofei Han Wenji Li Dongliang Wang Yindong Chen Zhifeng Hao Ruichu Cai Jiafan Zhuang |
| author_sort | Zhun Fan |
| collection | DOAJ |
| description | Deep reinforcement learning (DRL) has significantly advanced online path planning for unmanned aerial vehicles (UAVs). Nonetheless, DRL-based methods often encounter reduced performance when dealing with unfamiliar scenarios. This decline is mainly caused by the presence of non-causal and domain-specific elements within visual representations, which negatively impact the learning of policies. Present techniques generally depend on predefined augmentation or regularization methods intended to direct the model toward identifying causal and domain-invariant components, thereby enhancing the model’s ability to generalize. However, these manually crafted approaches are intrinsically constrained in their coverage and do not consider the entire spectrum of possible scenarios, resulting in less effective performance in new environments. Unlike prior studies, this work prioritizes representation learning and presents a novel method for causal representation disentanglement. The approach successfully distinguishes between causal and non-causal elements in visual data. By concentrating on causal aspects during the policy learning phase, the impact of non-causal factors is minimized, thereby improving the generalizability of DRL models. Experimental results demonstrate that our technique achieves reliable navigation and effective collision avoidance in unseen scenarios, surpassing state-of-the-art deep reinforcement learning algorithms. |
| format | Article |
| id | doaj-art-43ea04e4308342acb2a07eaeabe965bb |
| institution | Kabale University |
| issn | 2504-446X |
| language | English |
| publishDate | 2024-12-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Drones |
| spelling | doaj-art-43ea04e4308342acb2a07eaeabe965bb2025-01-24T13:29:38ZengMDPI AGDrones2504-446X2024-12-01911010.3390/drones9010010UAV Collision Avoidance in Unknown Scenarios with Causal Representation DisentanglementZhun Fan0Zihao Xia1Che Lin2Gaofei Han3Wenji Li4Dongliang Wang5Yindong Chen6Zhifeng Hao7Ruichu Cai8Jiafan Zhuang9Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Shenzhen 518110, ChinaCollege of Engineering, Shantou University, Shantou 515063, ChinaCollege of Engineering, Shantou University, Shantou 515063, ChinaCollege of Engineering, Shantou University, Shantou 515063, ChinaInternational Cooperation Base of Evolutionary Intelligence and Robotics, Shantou 515041, ChinaInternational Cooperation Base of Evolutionary Intelligence and Robotics, Shantou 515041, ChinaCollege of Engineering, Shantou University, Shantou 515063, ChinaCollege of Engineering, Shantou University, Shantou 515063, ChinaSchool of Computer Science, Guangdong University of Technology, Guangzhou 510006, ChinaInternational Cooperation Base of Evolutionary Intelligence and Robotics, Shantou 515041, ChinaDeep reinforcement learning (DRL) has significantly advanced online path planning for unmanned aerial vehicles (UAVs). Nonetheless, DRL-based methods often encounter reduced performance when dealing with unfamiliar scenarios. This decline is mainly caused by the presence of non-causal and domain-specific elements within visual representations, which negatively impact the learning of policies. Present techniques generally depend on predefined augmentation or regularization methods intended to direct the model toward identifying causal and domain-invariant components, thereby enhancing the model’s ability to generalize. However, these manually crafted approaches are intrinsically constrained in their coverage and do not consider the entire spectrum of possible scenarios, resulting in less effective performance in new environments. Unlike prior studies, this work prioritizes representation learning and presents a novel method for causal representation disentanglement. The approach successfully distinguishes between causal and non-causal elements in visual data. By concentrating on causal aspects during the policy learning phase, the impact of non-causal factors is minimized, thereby improving the generalizability of DRL models. Experimental results demonstrate that our technique achieves reliable navigation and effective collision avoidance in unseen scenarios, surpassing state-of-the-art deep reinforcement learning algorithms.https://www.mdpi.com/2504-446X/9/1/10deep reinforcement learningunmanned aerial vehiclesgeneralization capabilitypolicy learningcausal representation disentanglement |
| spellingShingle | Zhun Fan Zihao Xia Che Lin Gaofei Han Wenji Li Dongliang Wang Yindong Chen Zhifeng Hao Ruichu Cai Jiafan Zhuang UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement Drones deep reinforcement learning unmanned aerial vehicles generalization capability policy learning causal representation disentanglement |
| title | UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement |
| title_full | UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement |
| title_fullStr | UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement |
| title_full_unstemmed | UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement |
| title_short | UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement |
| title_sort | uav collision avoidance in unknown scenarios with causal representation disentanglement |
| topic | deep reinforcement learning unmanned aerial vehicles generalization capability policy learning causal representation disentanglement |
| url | https://www.mdpi.com/2504-446X/9/1/10 |
| work_keys_str_mv | AT zhunfan uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT zihaoxia uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT chelin uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT gaofeihan uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT wenjili uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT dongliangwang uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT yindongchen uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT zhifenghao uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT ruichucai uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement AT jiafanzhuang uavcollisionavoidanceinunknownscenarioswithcausalrepresentationdisentanglement |