Multi-sensor fusion, motor control and localization for biomimetic robotic dogs

Aiming at the localization and motor control challenges of biomimetic robotic dogs in complex environments, this study proposes a technical scheme integrating multi-sensor fusion, localization systems, and motor control. A visual-inertial-leg (VIO-LIO) multi-sensor fusion localization system is established: dynamic local feature extraction optimizes the visual module, while leg-inertial odometry is built using motor encoders and IMU, and Unscented Kalman Filter (UKF) fusion reduces localization error by over 20% and data processing time by 15.8%. Additionally, fused localization data is integrated with an improved Model Predictive Control (MPC) algorithm; a Convolutional Neural Network (CNN) predicts gait data to adjust motor joint parameters in real time through feedback, enhancing motor control stability. Experiments verify that in scenarios like slippery surfaces and dim environments, the system achieves a localization RMSE as low as 0.0352m, and motor-driven gait stability is 12% higher than traditional methods, providing reliable support for biomimetic robotic dogs’ complex-scene applications.