: First presenter name: Ori Barak subject : learning the myopic policy in restless bandits with unknown markovian dynamics : Second presenter name : Hod Yehuda Almakayes Seminar Topic: “Robot Gripper Pose Estimation using a Single Monocular Camera”

name: Ori Barak degree : third supervisor : prof. Kobi cohen subject : learning the myopic policy in restless bandits with unknown markovian dynamics abstract: We study the restless multi-armed bandit (RMAB) problem with unknown Markovian dynamics, where only the state of the selected arm is observed while all arms evolve over time. We introduce the Myopic Learning for Restless Bandits (MLRB) algorithm, which learns to act according to the myopic policy—a dynamic strategy that selects the arm with the highest expected immediate reward based on belief updates. Unlike prior works that measure weak regret against a fixed arm, we define a stronger myopic-genie regret, comparing performance to a genie that knows the true transition probabilities and follows the optimal myopic policy. We prove that MLRB achieves logarithmic regret, matching the best achievable order, and validate the results through simulations showing that MLRB significantly outperforms state-of-the-art RMAB learning algorithms. Hod Yehuda Almakayes Ms.C student Supervisor: Dr. Igal Bilik Seminar Topic: “Robot Gripper Pose Estimation using a Single Monocular Camera” Abstract: Human safety is critical for integrating autonomous robots into human living environments. Ensuring safety requires accurate, reliable, and real-time robot localization. The research focuses on robot gripper localization using a low-cost single monocular RGB camera with known intrinsic parameters. The proposed approach, denoted as frame-by-frame integrated network and geometry for gripper pose estimation (FINE-GPE). A comprehensive dataset of robot arm motions was generated under various conditions, including clutter and occlusions. Experimental results demonstrate that our approach is robust to robot arm occlusions and background clutter, showing low sensitivity to specific camera types.