Enhancing automated robotic services efficiency through intelligent dependent robotic computation

This research focuses on enhancing the adaptability and efficacy of service robots in real-time, multi-scenario environments where diverse settings complicate the interpretation of user commands and dynamic environmental fluctuations. It introduces the Commuting Input Valuation Approach (CIVA), which combines transfer learning with a flexible state transfer system to improve robot response rates, minimize unnecessary actions, and enhance learning efficiency across varied environments. The main contributions are (i) a continuous training framework for robots, (ii) a method for sharing information between response and learning stages, (iii) transfer learning approaches suited for both short and long inputs, and (iv) an adaptive reaction rate calculation that accounts for real-time conditions. CIVA was evaluated on 32 interactive tasks using the Daily Interactive Robot Manipulation (DIM) dataset, containing 1,603 dependent and 1,751 independent commands. Performance was assessed relative to baseline robotic models utilizing criteria including task completion duration, input interpreting failure rate, command-to-action efficiency, reaction successful rate, and learning velocity. In a targeted screw-loosening assessment comprising 120 input commands over 70 seconds, CIVA demonstrated a 35% reduction in task completion duration, a 42% decline in interpretation errors, a 28% enhancement in instruction-to-action efficiency, a 45% augmentation in response success rate, and a 50% increase in learning velocity relative to baseline measurements. The findings indicate that CIVA can enhance human-robot interaction and task performance; nevertheless, additional validation is necessary to verify reproducibility and generalization across diverse real-world contexts.