Improved YOLOv8 and IMU-UWB Fusion for Enhanced Target Detection and Precise Positioning in Quadruped Robots

To address the challenges of target detection and precise positioning for quadruped robots in complex environments, this paper proposes a method that combines an improved YOLOv8 model with IMU-UWB integrated positioning technology. In terms of target detection, the YOLOv8 model was optimized by replacing the original backbone with the lightweight ShuffleNetV2 and removing certain convolutional modules, reducing parameters by 43% (from 11.17M to 6.39M) to enhance efficiency on embedded devices. While mAP50-95 dropped to 60.35%. The mAP50-95 dropped to 60.35%, which lead to a slight decrease of 7% compared to the original model. The model maintained high accuracy under less strict conditions with a mAP50 of 93.4%, comparable to the original's 93.8%. This optimization provides a balance between reduced computational complexity and robust detection performance for embedded applications. In terms of positioning technology, this paper integrates IMU and UWB technologies, utilizing an Extended Kalman Filter (EKF) to achieve multi-sensor data fusion. This method fully leverages the dynamic response characteristics of IMU and the high-precision positioning capability of UWB, reducing positioning errors to within 0.5 meters, significantly enhancing the system's stability and robustness. The combination of these two methods greatly improves the perception capability and autonomy of quadruped robots in complex scenarios, demonstrating broad potential for applications in fields such as drone collaboration and autonomous driving.