2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

6D Pose Estimation for Flexible Production with Small Lot Sizes based on CAD Models using Gaussian Process Implicit Surfaces

Jianjie Lin, Markus Rickert, Alois Knoll
Robotics, Artificial Intelligence and Real-time Systems
Department of Informatics, Technische Universität München

DOI preprint BibTeX Code

Abstract

We propose a surface-to-surface (S2S) point registration algorithm by exploiting the Gaussian Process Implicit Surfaces for partially overlapping 3D surfaces to estimate the 6D pose transformation. Unlike traditional approaches, that separate the corresponding search and update steps in the inner loop, we formulate the point registration as a nonlinear non- constraints optimization problem which does not explicitly use any corresponding points between two point sets. According to the implicit function theorem, we form one point set as a Gaussian Process Implicit Surfaces utilizing the signed distance function, which implicitly creates three manifolds. Points on the same manifold share the same function value, indicated as {𝟏, 𝟎, −𝟏}. The problem is thus converted into finding a rigid transformation that minimizes the inherent function value. This can be solved by using a Gauss-Newton (GN) or Levenberg- Marquardt (LM) solver. In the case of a partially overlapping 3D surface, the Fast Point Feature Histogram (FPFH) algorithm is applied to both point sets and a Principal Component Anal- ysis (PCA) is performed on the result. Based on this, the initial transformation can then be computed. We conduct experiments on multiple point sets to evaluate the effectiveness of our proposed approach against existing state-of-the-art methods.

Qualitative comparison on Standard Bunny

Stanford Bunny together with alignment results of selected algorithms for (a)--(f) 50° rotation around the y axis and small translation, (g)--(i) 180° rotation around the z axis with translation set to zero.

Qualitative comparison

Benchmark results for all algorithms on four different point sets with three levels of Gaussian noise and three different overlap factors. The best RMSE value ε for each configuration is highlighted in bold

noise=0.00000 noise=0.00025 noise=0.00050
1.00 0.85 0.65 1.00 0.85 0.65 1.00 0.85 0.65
ε t ε t ε t ε t ε t ε t ε t ε t ε t
Bunny PCL-ICP 0.081 0.55 0.090 0.39 0.067 0.33 0.046 0.42 0.082 0.64 0.046 0.31 0.069 0.38 0.107 0.62 0.073 0.31
GoICP 0.115 20.29 0.110 20.06 0.097 19.98 0.050 20.15 0.020 20.06 0.060 20.17 0.090 20.08 0.100 20.07 0.050 20.13
GoICPT 0.100 21.49 0.110 21.47 0.100 21.36 0.110 21.51 0.100 21.72 0.100 21.43 0.100 21.43 0.110 21.68 0.100 21.43
SAC-IA-ICP 0.001 6.14 0.002 5.63 0.010 7.12 0.001 6.63 0.002 6.05 0.010 8.24 0.001 7.02 0.003 6.41 0.011 7.92
Gl.RANSAC 0.001 1.599 0.005 1.678 0.005 2.312 0.001 1.699 0.005 1.839 0.005 1.948 0.001 1.710 0.005 1.872 0.005 2.348
FGR 0.017 0.400 0.009 0.370 0.017 0.322 0.004 0.385 0.007 0.373 0.020 0.348 0.004 0.412 0.007 0.388 0.015 0.348
Ours 0.001 0.568 0.001 0.486 0.001 0.465 0.001 0.531 0.001 0.609 0.002 0.507 0.001 0.466 0.002 0.561 0.002 0.690
Suzanne PCL-ICP 0.134 0.810 0.108 0.814 0.116 0.656 0.049 0.587 0.127 1.264 0.095 0.650 0.131 0.934 0.115 0.619 0.106 0.902
GoICP 0.134 0.810 0.108 0.814 0.116 0.656 0.049 0.587 0.127 1.264 0.095 0.650 0.131 0.934 0.115 0.619 0.106 0.902
GoICPT 0.092 23.150 0.059 21.502 0.084 21.532 0.079 21.584 0.071 21.679 0.097 21.655 0.045 21.517 0.091 21.569 0.062 21.473
SAC-IA-ICP 0.092 23.150 0.059 21.502 0.084 21.532 0.079 21.584 0.071 21.679 0.097 21.655 0.045 21.517 0.091 21.569 0.062 21.473
Gl.RANSAC 0.014 1.663 0.018 1.724 0.040 2.002 0.016 1.829 0.011 1.808 0.039 2.277 0.013 1.865 0.025 1.912 0.022 2.711
FGR 0.049 0.583 0.039 0.574 0.060 0.512 0.070 0.628 0.049 0.602 0.061 0.532 0.046 0.697 0.071 0.641 0.052 0.572
Ours 0.049 0.583 0.039 0.574 0.060 0.512 0.070 0.628 0.049 0.602 0.061 0.532 0.046 0.697 0.071 0.641 0.052 0.572
Dragon PCL-ICP 0.087 0.487 0.100 0.317 0.079 0.445 0.090 0.229 0.065 0.346 0.096 0.352 0.071 0.502 0.082 0.465 0.101 0.382
GoICP 0.017 21.581 0.022 21.736 0.018 21.416 0.034 21.490 0.035 21.697 0.021 21.497 0.018 21.474 0.053 21.467 0.033 21.562
GoICPT 0.022 21.388 0.014 21.370 0.021 20.017 0.011 19.945 0.009 19.976 0.084 20.054 0.013 20.136 0.044 20.168 0.017 20.170
SAC-IA-ICP 0.001 5.927 0.003 5.279 0.009 4.795 0.001 6.018 0.003 5.433 0.009 4.852 0.001 6.478 0.004 5.816 0.009 5.215
Gl.RANSAC 0.001 2.078 0.005 2.555 0.005 2.867 0.001 2.274 0.005 2.468 0.004 2.832 0.001 2.582 0.005 2.493 0.005 2.812
FGR 0.001 2.078 0.005 2.555 0.005 2.867 0.001 2.274 0.005 2.468 0.004 2.832 0.001 2.582 0.005 2.493 0.005 2.812
Ours 0.012 0.501 0.022 0.455 0.024 0.391 0.012 0.478 0.016 0.444 0.017 0.382 0.021 0.499 0.015 0.472 0.013 0.395
Buddha PCL-ICP 0.002 0.705 0.002 0.950 0.002 1.017 0.002 0.848 0.002 0.920 0.003 1.016 0.002 0.740 0.002 0.962 0.003 0.872
GoICP 0.043 21.539 0.032 21.401 0.085 21.309 0.075 21.380 0.050 21.295 0.052 21.341 0.051 21.368 0.067 21.474 0.012 21.390
GoICPT 0.013 20.168 0.025 20.133 0.028 20.171 0.016 20.167 0.031 20.191 0.022 20.115 0.023 19.983 0.031 19.910 0.020 19.965
SAC-IA-ICP 0.001 5.534 0.007 5.216 0.009 6.265 0.001 6.179 0.014 7.343 0.017 7.428 0.001 6.700 0.011 7.548 0.015 6.169
Gl.RANSAC 0.001 5.534 0.007 5.216 0.009 6.265 0.001 6.179 0.014 7.343 0.017 7.428 0.001 6.700 0.011 7.548 0.015 6.169
FGR 0.022 0.393 0.022 0.363 0.019 0.327 0.016 0.397 0.019 0.380 0.015 0.323 0.024 0.421 0.022 0.393 0.021 0.338
Ours 0.003 0.989 0.002 0.768 0.002 0.844 0.001 0.581 0.003 0.709 0.002 0.830 0.002 0.681 0.002 0.739 0.003 1.165

Reference

For more detailed information please have a look at our IROS paper. The reference is

Jianjie Lin, Markus Rickert, Alois Knoll. 6D Pose Estimation for Flexible Production with Small Lot Sizes based on CAD Models using Gaussian Process Implicit Surfaces, In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Las Vegas, NV, USA, Oct. 2020.

@INPROCEEDINGS{lingpis2020,
    author={Lin, Jianjie and Rickert, Markus and Knoll, Alois},
    booktitle={2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, 
    title={6D Pose Estimation for Flexible Production with Small Lot Sizes based on CAD Models using Gaussian Process Implicit Surfaces}, 
    year={2020},
    volume={},
    number={},
    pages={10572-10579},
    doi={10.1109/IROS45743.2020.9341189}}