Deep Source-Seekers with Obstacle Avoidance: Adaptive Hybrid Control with Transformers In-The-Loop

Xiyuan Zhang, Daniel Ochoa, Regina Talonia, Jorge Poveda
Proceedings of the 7th Annual Learning for Dynamics \& Control Conference, PMLR 283:844-855, 2025.

Abstract

Autonomous signal source localization is a cornerstone of modern robotics, underpinning critical applications in environmental monitoring, search and rescue, and industrial automation. Traditional source-seeking methods, such as gradient-based algorithms and potential field-based approaches, often struggle with local minimum entrapment in environments cluttered with obstacles. To address these challenges, in this paper we introduce a novel model-free approach that combines a perception-driven hybrid controller—integrating adaptive continuous-time and discrete-time feedback—with an Environmental Complexity Adapter (ECA) for perception model selection. The proposed dynamics implement real-time exploration/exploitation mechanisms and complementary deep learning-based perception architectures: YOLOv10 for rapid and accurate object detection in clear conditions, and Real-Time DEtection TRansformer (RT-DETR) for enhanced robustness in noisy environments. By continuously assessing the quality of sensor data, the ECA dynamically switches between these models, optimizing the trade-off between processing speed and detection reliability. This approach harnesses the robustness of hybrid controllers while enabling efficient, perception-guided source-seeking and obstacle avoidance in complex environments. Extensive numerical simulations validate the effectiveness of the proposed approach.

Cite this Paper

BibTeX
@InProceedings{pmlr-v283-zhang25a, title = {Deep Source-Seekers with Obstacle Avoidance: Adaptive Hybrid Control with Transformers In-The-Loop}, author = {Zhang, Xiyuan and Ochoa, Daniel and Talonia, Regina and Poveda, Jorge}, booktitle = {Proceedings of the 7th Annual Learning for Dynamics \& Control Conference}, pages = {844--855}, year = {2025}, editor = {Ozay, Necmiye and Balzano, Laura and Panagou, Dimitra and Abate, Alessandro}, volume = {283}, series = {Proceedings of Machine Learning Research}, month = {04--06 Jun}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v283/main/assets/zhang25a/zhang25a.pdf}, url = {https://proceedings.mlr.press/v283/zhang25a.html}, abstract = {Autonomous signal source localization is a cornerstone of modern robotics, underpinning critical applications in environmental monitoring, search and rescue, and industrial automation. Traditional source-seeking methods, such as gradient-based algorithms and potential field-based approaches, often struggle with local minimum entrapment in environments cluttered with obstacles. To address these challenges, in this paper we introduce a novel model-free approach that combines a perception-driven hybrid controller—integrating adaptive continuous-time and discrete-time feedback—with an Environmental Complexity Adapter (ECA) for perception model selection. The proposed dynamics implement real-time exploration/exploitation mechanisms and complementary deep learning-based perception architectures: YOLOv10 for rapid and accurate object detection in clear conditions, and Real-Time DEtection TRansformer (RT-DETR) for enhanced robustness in noisy environments. By continuously assessing the quality of sensor data, the ECA dynamically switches between these models, optimizing the trade-off between processing speed and detection reliability. This approach harnesses the robustness of hybrid controllers while enabling efficient, perception-guided source-seeking and obstacle avoidance in complex environments. Extensive numerical simulations validate the effectiveness of the proposed approach.} }
Endnote
%0 Conference Paper %T Deep Source-Seekers with Obstacle Avoidance: Adaptive Hybrid Control with Transformers In-The-Loop %A Xiyuan Zhang %A Daniel Ochoa %A Regina Talonia %A Jorge Poveda %B Proceedings of the 7th Annual Learning for Dynamics \& Control Conference %C Proceedings of Machine Learning Research %D 2025 %E Necmiye Ozay %E Laura Balzano %E Dimitra Panagou %E Alessandro Abate %F pmlr-v283-zhang25a %I PMLR %P 844--855 %U https://proceedings.mlr.press/v283/zhang25a.html %V 283 %X Autonomous signal source localization is a cornerstone of modern robotics, underpinning critical applications in environmental monitoring, search and rescue, and industrial automation. Traditional source-seeking methods, such as gradient-based algorithms and potential field-based approaches, often struggle with local minimum entrapment in environments cluttered with obstacles. To address these challenges, in this paper we introduce a novel model-free approach that combines a perception-driven hybrid controller—integrating adaptive continuous-time and discrete-time feedback—with an Environmental Complexity Adapter (ECA) for perception model selection. The proposed dynamics implement real-time exploration/exploitation mechanisms and complementary deep learning-based perception architectures: YOLOv10 for rapid and accurate object detection in clear conditions, and Real-Time DEtection TRansformer (RT-DETR) for enhanced robustness in noisy environments. By continuously assessing the quality of sensor data, the ECA dynamically switches between these models, optimizing the trade-off between processing speed and detection reliability. This approach harnesses the robustness of hybrid controllers while enabling efficient, perception-guided source-seeking and obstacle avoidance in complex environments. Extensive numerical simulations validate the effectiveness of the proposed approach.
APA
Zhang, X., Ochoa, D., Talonia, R. & Poveda, J.. (2025). Deep Source-Seekers with Obstacle Avoidance: Adaptive Hybrid Control with Transformers In-The-Loop. Proceedings of the 7th Annual Learning for Dynamics \& Control Conference, in Proceedings of Machine Learning Research 283:844-855 Available from https://proceedings.mlr.press/v283/zhang25a.html.

Related Material