YOPOv2-Tracker: An End-to-End Agile Tracking and Navigation Framework

Abstract

To track fast-moving targets in cluttered environments, a series of improvements in detection, mapping, navigation, and control has been introduced in previous work to make the overall system more comprehensive. However, this separated pipeline introduces significant latency and limits the agility of quadrotors. On the contrary, we follow the design principle of “less is more,” striving to simplify the process while maintaining effectiveness. In this work, we propose an end-to-end agile tracking and navigation framework for quadrotors with an elegant structure. Importantly, leveraging the multimodal nature of navigation and detection tasks, our network maintains interpretability by explicitly integrating the independent modules of the traditional pipeline, rather than a crude action regression. In detail, we adopt a set of motion primitives as anchors to cover the searching space regarding the feasible region and potential target. Then, we reformulate trajectory optimization as the regression of primitive offsets considering safety, smoothness, and other metrics. For the tracking task, the trajectories are expected to approach the target, and additional class scores are predicted. Subsequently, the predictions, after compensation for the estimated lumped disturbance, are transformed into thrust and attitude as control commands for swift response. We seamlessly integrate traditional planning with data-driven learning by computing the cost gradients with respect to the trajectory parameters, as in classical optimization, and directly backpropagating them to the network weights. This eliminates the need for expert demonstration in imitation learning and provides more direct guidance than reinforcement learning. Finally, we deploy the algorithm on a compact quadrotor and conduct real-world validations in both forest and building environments to demonstrate the efficiency of the proposed method.