用MPC提升Anafi无人机追踪精度，打通Sphinx与Gazebo仿真壁垒

Simulation frameworks are essential for the safe development of robotic applications. However, different components of a robotic system are often best simulated in different environments, making full integration challenging. This is particularly true for partially-open or closed-source simulators, which commonly suffer from two limitations: (i) lack of runtime control over scene actors via interfaces like ROS, and (ii) restricted access to real-time state data (e.g., pose, velocity) of scene objects. In the first part of this work, we address these issues by integrating aerial drones simulated in Parrot's Sphinx environment (used for Anafi drones) into the Gazebo simulator. Our approach uses a mirrored drone instance embedded within Gazebo environments to bridge the two simulators. One key application is aerial target tracking, a common task in multi-robot systems. However, Parrot's default PID-based controller lacks the agility needed for tracking fast-moving targets. To overcome this, in the second part of this work we develop a model predictive controller (MPC) that leverages cumulative error states to improve tracking accuracy. Our MPC significantly outperforms the built-in PID controller in dynamic scenarios, increasing the effectiveness of the overall system. We validate our integrated framework by incorporating the Anafi drone into an existing Gazebo-based airship simulation and rigorously test the MPC against a custom PID baseline in both simulated and real-world experiments.