Autonomous Robotic Stone Stacking with Online next Best Object Target Pose Planning

Goal: 用不规则形状的石头（已知的set，也有pre-scanned model）搭建一个vertical tower
Contribution:

1. pose searching algorithm（stacking algorithm）
   目的是在object set中找到best object 和它的target pose。
   search space: discrete object space and continuous pose space
   怎样评价object和相应pose的goodness: 引入了代价函数
   用物理引擎考虑搭建的结构稳定性（用物理引擎得到CoM，动力学特性以及不规则物体之间的接触点）
2. object detection pipeline
   场景中的物体都有pre-scanned model，所以object detection pipeline有如下steps:
   1.Extract 3D keypoints and Description :
   from raw point clouds of an RGB-D sensor, Rotational Projection Statistics (RoPS) descriptor has the best result.
   2.Descriptor matching and Clustering
   3.Transform refinement and Verification
   4.Object in Robot Arm Frame