RoboKeyGen: Robot Pose and Joint Angles Estimation via Diffusion-based 3D Keypoint Generation

• DOI: 10.48550/arxiv.2403.18259
• Publication Date: 2024-03-27
• Authors: Yang Tian; Jiyao Zhang; Guo-Wei Huang; Bin Wang; Ping Wang; Jiangmiao Pang; Hao Dong
• Keywords: FOS: Computer and information sciences; Computer Science - Robotics; Robotics (cs.RO)

Estimating robot pose and joint angles is significant in advanced robotics, enabling applications like collaboration online hand-eye calibration.However, the introduction of unknown makes prediction more complex than simple estimation, due to its higher dimensionality.Previous methods either regress 3D keypoints directly or utilise a render&compare strategy. These approaches often falter terms performance efficiency grapple with cross-camera gap problem.This paper presents novel framework that bifurcates high-dimensional task into two manageable subtasks: 2D detection lifting 3D. This separation promises enhanced without sacrificing innate keypoint-based techniques.A vital component our method keypoints. Common deterministic regression may when faced uncertainties from errors self-occlusions.Leveraging robust modeling potential diffusion models, we reframe this issue as conditional generation task. To bolster adaptability, introduce theNormalised Camera Coordinate Space (NCCS), ensuring alignment estimated across varying camera intrinsics.Experimental results demonstrate proposed outperforms state-of-the-art render\&compare achieves inference speed.Furthermore, tests accentuate method's generalisation capabilities.We intend release both dataset code https://nimolty.github.io/Robokeygen/