Neural Network-Based Ball Trajectory Control of Solenoid Kickers for Autonomous Soccer Robots

In the dynamic landscape of autonomous soccer, achieving precise and adaptive trajectory control for solenoid-based kicking systems remains a critical challenge. To address this, we propose a novel neural network-based control system, implemented on the IRIS autonomous soccer robot, that significantly enhances solenoid kicker accuracy and efficiency by eliminating manual calibration. Our approach employs a Multi-Layer Perceptron (MLP) to predict optimal kicker height from robot-togoal distance and capacitor discharge time, enabling dynamic, on-the-fly adjustments vital for rapid gameplay. Validated through experiments conducted in Lab robotics IRIS in 2024, our system demonstrated robust predictive performance, yielding a Mean Absolute Error (MAE) of 0.93 cm and a Root Mean Squared Error (RMSE) of $\mathbf{1. 3 4 ~ c m}$. Optimized for real-time deployment, the trained network was compiled into a compact Look-Up Table (LUT), drastically reducing the average prediction time to a mere 17.09 microseconds, a significant reduction in 44 % computation time over direct model inference. This makes it exceptionally well-suited for fast-paced robotic scenarios, offering superior precision, adaptability, and autonomy. Despite its high precision, the system's effectiveness is limited beyond 456 cm by the solenoid's physical energy constraints; future work will prioritize expanding this operational range to enhance performance and robustness across diverse field conditions.