Deep reinforcement learning-based thermal-visual collaborative optimization control system for multi-sensory art...

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Digital Art and Media

Background:

• Multi-sensory art installations require sophisticated control systems for immersive experiences.
• Traditional control methods struggle with dynamic, real-time optimization of thermal and visual elements.
• Integrating diverse sensory inputs presents significant fusion and processing challenges.

Purpose of the Study:

• To develop an attention-enhanced deep reinforcement learning (DRL) control system for optimizing thermal-visual collaboration in multi-sensory art installations.
• To improve control accuracy, reduce response times, and enhance energy efficiency compared to existing methods.
• To validate the system's practical effectiveness and user satisfaction in real-world deployments.

Main Methods:

• An attention-based Deep Deterministic Policy Gradient (DDPG) algorithm with dynamic modality weighting was employed.
• Adaptive sensor fusion techniques, combining Kalman and particle filtering, were implemented with low processing latency (8 ± 2 ms).
• A hierarchical four-layer architecture (perception, fusion, decision-making, execution) was designed for efficient control.

Main Results:

• Achieved a 65% improvement in control accuracy (0.085 ± 0.012 RMSE) and a 40% reduction in response time (45 ± 8 ms).
• Demonstrated a 23% reduction in energy consumption through intelligent thermal-visual coordination.
• Real-world deployment showed high user satisfaction (4.1/5) and system availability (98.5%).

Conclusions:

• The proposed attention-enhanced DRL system offers superior performance for thermal-visual collaborative optimization in art installations.
• The system enables more sophisticated and responsive multi-sensory experiences while maintaining artistic integrity.
• This research provides a foundation for advanced digital art technologies and intelligent interactive systems.