MPE: MultiPlex Interactive Learning for Multi-Robot Collaborative 6D Object Pose Estimation

In recent years, 6D object pose estimation has made significant strides. While state-of-the-art methods now generalize to objects beyond the training data, they often rely on multi-view labeled reference images, limiting their applicability in real-world scenarios. This paper introduces the multi-robot collaborative setting to the problem of pose estimation and designs a network called MultiPlex Interactive Learning for Multi-Robot Collaborative 6D Object Pose Estimation (MPE). It predicts the relative pose of an object in an image with respect to a reference image object, eliminating the need for 3D model reconstruction. Specifically, matched image pairs and keypoints are extracted using a pre-trained vision model, followed by interactive learning through the proposed MultiPlex Interactive Learning strategy. This end-to-end approach directly regresses 6DoF object poses, replacing the traditional PnP algorithm and ensuring robust full 6DoF pose estimation. Experiments on Linemod and YCB-Video datasets demonstrate that MPE achieves state-of-the-art performance across multiple evaluation metrics, delivering superior accuracy and broader applicability.