6DoF Object Pose Estimation

Keypoint voting methods for RGB-D, point cloud, and self-supervised pose estimation.

Research from my PhD at Queen’s University on 6DoF object pose estimation for robot perception. This line of work includes RCVPose, RCVPose3D, KeyGNet, and RKHSPose.

Core topics:

  • Radial keypoint voting for RGB-D 6DoF pose estimation.
  • Cascaded keypoint voting for point-cloud-only pose estimation.
  • Learned keypoint selection for multi-object settings.
  • Self-supervised sim-to-real pose estimation with pseudo-keypoints and RKHS learning.

Project links: RCVPose code, RCVPose paper, RCVPose3D paper, RKHSPose paper.

References