Master’s Thesis: End-to-End RL for Robust Teleoperation under Stochastic Delays
ImageClassification
Designed and trained deep learning models using LSTM and SAC on real-time noisy sensor image data. Built a complete data pipeline from data preprocessing, supervised initialization, RL fine-tuning, to systematic evaluation and deployment on embedded hardware. Utilized domain randomization to improve generalization of the labeled datasets for robust task performance.
• Labeled and processed image data to enable end-to-end RL model training.
• Implemented domain randomization to enhance training robustness.
• Verified labeled data and model outputs on physical hardware.
• Achieved performance improvement over state-of-the-art baselines.
Designed and trained deep learning models using LSTM and SAC on real-time noisy sensor image data. Built a complete data pipeline from data preprocessing, supervised initialization, RL fine-tuning, to systematic evaluation and deployment on embedded hardware. Utilized domain randomization to improve generalization of the labeled datasets for robust task performance. • Labeled and processed image data to enable end-to-end RL model training. • Implemented domain randomization to enhance training robustness. • Verified labeled data and model outputs on physical hardware. • Achieved performance improvement over state-of-the-art baselines.
2025 - 2026