Giri Wahyu Wiriasto, Siti Sendari, Dyah Lestari, Muhamad Syamsu Iqbal, Norrima Mokhtar
Energy efficiency is vital for modern robotic manipulators. This study proposes an Energy-Aware Q-Learning framework, using a reward system to balance power use and motion performance for a single-joint manipulator. Five energy-to-performance reward weight configurations were tested: (80:20), (70:30), (60:40), (50:50), and (40:60). Results show the 70:30 ratio achieved the lowest energy consumption (0.075 J). However, the 60:40 configuration provided the optimal trade-off, delivering the highest efficiency score (662.25 L1), highest average reward (875.60), and a 100% task success rate. Excessive energy weights cause conservative behavior, while a moderate balance produces adaptive-stable control, proving that energy-based rewards enhance the adaptability and efficiency of reinforcement learning control. © The 2026 International Conference on Artificial Life and Robotics (ICAROB2026).
Department of Electrical and Informatics Engineering, Universitas Negeri, Malang, 65145, Indonesia; Department of Electrical Engineering, Universitas Mataram, 83115, Indonesia; Department of Electrical Engineering, Faculty of Engineering, Universiti Malaya, Kuala Lumpur, Lembah Pantai, Malaysia
