6-DoF Object Pose Estimation for Robotic Manipulation

Duration: TBD
Affiliation: VinMotion, VinGroup · Hanoi, Vietnam

Note: Full write-up coming soon.

Overview

Research and development of 6 Degrees-of-Freedom (6-DoF) object pose estimation for robotic manipulation. The work progresses through three generations of methods: hand-crafted pipelines, foundation model-based (FoundationPose), and model-free approaches for unseen objects.

Method Progression

1. Manual Pipeline — Feature-based matching + ICP refinement for known object CAD models
2. Foundation Model-Based — Adapting FoundationPose for in-the-wild robotic scenarios; fine-tuning and domain adaptation experiments
3. Model-Free Methods — Zero-shot pose estimation for unseen objects without CAD priors

Technologies

Python PyTorch FoundationPose Open3D ROS 2 Computer Vision 3D Perception