Barkin Dagda
Research Subjects
Barkin Dagda has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
Saber Fallah
3 shared publicationsVinal Asodia
3 shared publications- Muhammad Awais
University of Surrey
1 shared publication - Mustafa Yildirim
Connected and Autonomous Vehicles Lab, Department of Mechanical Engineering, University of Surrey, Guildford GU2 7XH, UK
1 shared publication - Yinglong He
University of Surrey
1 shared publication Zhenhua Feng
1 shared publication
Latest Publications
- Open Access
- 19 March 2026
- Published by Elsevier in SSRN Electronic Journal
Offline Reinforcement Learning using Human-Aligned Reward Labeling for Autonomous Emergency Braking in Occluded Pedestrian Crossing
Effective leveraging of real-world driving datasets is crucial for enhancing the training of autonomous driving systems. While Offline Reinforcement Learning enables training autonomous vehicles with such data, most available datasets lack meaningful reward labels. Reward labeling is essential as it provides feedback for the learning algorithm to distinguish between desirable and undesirable behaviors, thereby improving policy performance. This paper presents a novel approach for generating human-aligned reward labels. The proposed approach addresses the challenge of absent reward signals in the real-world datasets by generating labels that reflect human judgment and safety considerations. The reward function incorporates an adaptive safety component that is activated by analyzing semantic segmentation maps, enabling the autonomous vehicle to prioritize safety over efficiency in potential collision scenarios. The proposed method is applied to an occluded pedestrian crossing scenario with varying pedestrian traffic levels, using simulation data. When the generated rewards were used to train various Offline Reinforcement Learning algorithms, each model produced a meaningful policy, demonstrating the method's viability. In addition, the method was applied to a subset of the Audi Autonomous Driving Dataset, and the reward labels were compared to human-annotated reward labels. The findings show a moderate disparity between the two reward sets, and, most interestingly, the method flagged unsafe states that the human annotator missed.
- 19 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessOffline Reinforcement Learning using Human-Aligned Reward Labeling for Autonomous Emergency Braking in Occluded Pedestrian Crossing
Effective leveraging of real-world driving datasets is crucial for enhancing the training of autonomous driving systems. While Offline Reinforcement Learning enables training autonomous vehicles with such data, most available datasets lack meaningful reward labels. Reward labeling is essential as it provides feedback for the learning algorithm to distinguish between desirable and undesirable behaviors, thereby improving policy performance. This paper presents a novel approach for generating human-aligned reward labels. The proposed approach addresses the challenge of absent reward signals in the real-world datasets by generating labels that reflect human judgment and safety considerations. The reward function incorporates an adaptive safety component that is activated by analyzing semantic segmentation maps, enabling the autonomous vehicle to prioritize safety over efficiency in potential collision scenarios. The proposed method is applied to an occluded pedestrian crossing scenario with varying pedestrian traffic levels, using simulation data. When the generated rewards were used to train various Offline Reinforcement Learning algorithms, each model produced a meaningful policy, demonstrating the method's viability. In addition, the method was applied to a subset of the Audi Autonomous Driving Dataset, and the reward labels were compared to human-annotated reward labels. The findings show a moderate disparity between the two reward sets, and, most interestingly, the method flagged unsafe states that the human annotator missed.
- Open Access
- 5 March 2026
- Published by Elsevier in SSRN Electronic Journal
VLM-Guided Reinforcement Learning for Emergency Braking in Occluded Pedestrian Crossing
Designing reward functions that remain reliable across diverse and dynamic driving scenarios remains a core challenge in reinforcement learning (RL) for autonomous vehicles. Conventional hand-crafted rewards are typically rigid, require extensive manual tuning, and often fail to capture deeper semantic understanding of the scene. This paper introduces a Vision–Language Model (VLM)–guided RL method for improving the adaptability and contextual awareness of RL policies, evaluated in autonomous emergency braking under occluded pedestrian scenarios. The proposed method integrates language-grounded scene understanding into both reward shaping and policy execution. During training, VLM-derived semantic feedback produces adaptive and context-aware reward signals that accelerate learning and promote safer behavior. However, as reward shaping is inherently limited to the training distribution, the learned policy may exhibit performance degradation under distribution shift. To address this limitation, the paper further introduces an inference-time VLM-guided policy refinement mechanism that incorporates zero-shot visual–semantic reasoning during the deployment process. Experiments in CARLA show improved collision avoidance, success rate, stopping distance, and convergence, with VLM-guided reward shaping and policy execution compared to fixed reward baselines.
- 5 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessVLM-Guided Reinforcement Learning for Emergency Braking in Occluded Pedestrian Crossing
Designing reward functions that remain reliable across diverse and dynamic driving scenarios remains a core challenge in reinforcement learning (RL) for autonomous vehicles. Conventional hand-crafted rewards are typically rigid, require extensive manual tuning, and often fail to capture deeper semantic understanding of the scene. This paper introduces a Vision–Language Model (VLM)–guided RL method for improving the adaptability and contextual awareness of RL policies, evaluated in autonomous emergency braking under occluded pedestrian scenarios. The proposed method integrates language-grounded scene understanding into both reward shaping and policy execution. During training, VLM-derived semantic feedback produces adaptive and context-aware reward signals that accelerate learning and promote safer behavior. However, as reward shaping is inherently limited to the training distribution, the learned policy may exhibit performance degradation under distribution shift. To address this limitation, the paper further introduces an inference-time VLM-guided policy refinement mechanism that incorporates zero-shot visual–semantic reasoning during the deployment process. Experiments in CARLA show improved collision avoidance, success rate, stopping distance, and convergence, with VLM-guided reward shaping and policy execution compared to fixed reward baselines.
- 1 July 2025
- Published by Elsevier in Robotics and Autonomous Systems
- Vol. 193
HighwayLLM: Decision-making and navigation in highway driving with RL-informed language model
- 1 July 2025
- Published by Elsevier in Robotics and Autonomous Systems
- Vol. 193
HighwayLLM: Decision-making and navigation in highway driving with RL-informed language model