advanced techniques for mobile robotics gaussian...
TRANSCRIPT
![Page 1: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/1.jpg)
Wolfram Burgard, Cyrill Stachniss,
Kai Arras, Maren Bennewitz
Gaussian Processes in Robotics
Advanced Techniques for Mobile Robotics
![Page 2: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/2.jpg)
Overview
§ Regression problem
§ Gaussian process models
§ Learning GPs
§ Applications
§ Summary
![Page 3: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/3.jpg)
The Regression Problem § Given n observed points
§ Assuming the dependency
§ How to predict new points
![Page 4: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/4.jpg)
The Regression Problem § Given n observed points
![Page 5: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/5.jpg)
The Regression Problem § Solution 1: Parametric models
§ Linear
§ Quadratic
§ Higher order polynomials
§ …
§ Learning: optimizing the parameters
![Page 6: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/6.jpg)
The Regression Problem § Solution 1: Parametric models
![Page 7: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/7.jpg)
The Regression Problem § Solution 1: Parametric models
![Page 8: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/8.jpg)
The Regression Problem § Solution 2: Non-parametric models
§ Radial Basis functions
§ Histograms, Splines, Support Vector Machines …
§ Learning: finding the structure of the model and optimize its parameters
![Page 9: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/9.jpg)
The Regression Problem § Given n observed points
![Page 10: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/10.jpg)
The Regression Problem § Solution 3: Express
directly in terms of the data points
§ Idea: Any finite set of values sampled from has a joint Gaussian distribution with a covariance matrix given by
![Page 11: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/11.jpg)
Gaussian Process Models § Then, the n+1 dimensional vector
which includes the new target to be predicted , comes from an n+1 dimensional Gaussian
§ The predictive distribution for the new target is a 1-dimensional Gaussian
![Page 12: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/12.jpg)
§ Given the n observed points § Squared exponential covariance
function
§ with § and a noise level
Gaussian Process Model
![Page 13: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/13.jpg)
The Regression Problem § Given n observed points
![Page 14: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/14.jpg)
Gaussian Process Models § GP model
![Page 15: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/15.jpg)
Learning GPs § The squared exponential
covariance function:
§ Easy to interpret parameters
amplitude
index/input distance
characteristic lengthscale
noise level
![Page 16: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/16.jpg)
Learning GPs § Example: low noise
![Page 17: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/17.jpg)
Learning GPs § Example: medium noise
![Page 18: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/18.jpg)
Learning GPs § Example: high noise
![Page 19: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/19.jpg)
Learning GPs § Example: small lengthscale
![Page 20: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/20.jpg)
Learning GPs § Example: large lengthscale
![Page 21: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/21.jpg)
Learning GPs § Covariance function specifies the prior
prior posterior
![Page 22: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/22.jpg)
Gaussian Process Models § Recall, the n+1 dimensional vector
comes from an n+1 dimensional normal distribution
§ The predictive distribution for the new target is a 1-dimensional Gaussian.
§ Why?
![Page 23: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/23.jpg)
The Gaussian Distribution § Recall the 2-dimensional joint Gaussian:
§ The conditionals and the marginals are also Gaussians Figure taken from
Carl E. Rasmussen: NIPS 2006 Tutorial
![Page 24: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/24.jpg)
The Gaussian Distribution § Simple bivariate example:
![Page 25: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/25.jpg)
The Gaussian Distribution § Simple bivariate example:
joint
marginal conditional
![Page 26: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/26.jpg)
The Gaussian Distribution § Marginalization:
![Page 27: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/27.jpg)
The Gaussian Distribution § The conditional:
![Page 28: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/28.jpg)
The Gaussian Distribution § Slightly more complicated in the general
case:
§ The conditionals and the marginals are also Gaussians
Figure taken from Carl E. Rasmussen: NIPS 2006 Tutorial
![Page 29: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/29.jpg)
The Gaussian Distribution § Conditioning the joint Gaussian in general
§ In case of zero mean:
![Page 30: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/30.jpg)
Gaussian Process Models § Recall the GP assumption
![Page 31: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/31.jpg)
Gaussian Process Models § Noise-free mean and variance of the
predictive distribution have the form
§ with
![Page 32: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/32.jpg)
Gaussian Process Models § Mean and variance of the predictive
distribution then lead to
§ with
![Page 33: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/33.jpg)
Learning GPs § Learning a Gaussian process means
§ choosing a covariance function § finding its parameters and the noise level
§ What is the objective?
![Page 34: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/34.jpg)
Learning GPs § The hyperparameters
can be found by maximizing the likelihood of the training data e.g., using gradient methods
![Page 35: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/35.jpg)
Learning GPs § Objective: high data likelihood
§ Due to the Gaussian assumption, GPs have Occam’s razor built in
data fit complexity penalty
const.
![Page 36: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/36.jpg)
Occam‘s Razor § Use the simplest explanation that is
needed to describe the data
§ Data-fit favors overfitting § Complexity penalty favors simplicity
too long just right too short
![Page 37: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/37.jpg)
Advanced Topics / Extensions § Classification/non-Gaussian noise § Sparse GPs: Approximations for large
data sets § Heteroscedastic GPs: Modeling non-
constant noise § Nonstationary GPs: Modeling varying
smoothness (lengthscales) § Mixtures of GPs § Uncertain inputs § …
![Page 38: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/38.jpg)
Further Reading
![Page 39: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/39.jpg)
Applications in Robotics § Monocular range sensing § Terrain modeling § Learning sensor models § Learning to control a blimp § Localization in cellular networks § Time-series forecasting § …
![Page 40: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/40.jpg)
Applications in Robotics § Monocular range sensing § Terrain modeling § Learning sensor models § Learning to control a blimp § Localization in cellular networks § Time-series forecasting § …
![Page 41: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/41.jpg)
Monocular Range Sensing
§ Can we learn range from single, monocular camera images?
![Page 42: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/42.jpg)
Training Setup § Mobile robot + laser range finder § Omni-directional monocular camera
![Page 43: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/43.jpg)
Training Setup DFKI Saarbrücken University of Freiburg
![Page 44: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/44.jpg)
Learning Range from Vision § Associate (polar) pixel columns with ranges
Extract features
Associate with ranges
![Page 45: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/45.jpg)
Pre-processing § Warp images into a panoramic view
§ 120 pixels per column
§ Transform to HSV -> 420 dimensions
![Page 46: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/46.jpg)
Visual Features § Two types of features
1. No human engineering: Principle components analysis (PCA) on raw input
2. Use of domain specific knowledge: Edge features that shall correspond to floor boundaries
![Page 47: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/47.jpg)
Experiments
Typical 180° scan
![Page 48: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/48.jpg)
Online Prediction
![Page 49: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/49.jpg)
Mapping Results Laser-based Vision-based
Saarbrücken:
Freiburg:
![Page 50: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/50.jpg)
GP-based Terrain Modeling § 3D terrain models are important in
many tasks in outdoor robotics
![Page 51: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/51.jpg)
Terrain Modeling § Given: observations of the terrain
surface § Task: Learn a predictive model § Classic Approach: Elevation grid maps
![Page 52: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/52.jpg)
GP-Based Approach § Generalize the grid-based model to
fully continuous spaces by viewing the problem as function regression
§ Requirements § Probabilistic formulation to handle
uncertainty § Ability to adapt to the spatial
structures
![Page 53: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/53.jpg)
Covariance Function § Standard covariance function have
limited flexibility to adapt to the local spatial structure
strong smoothing
medium smoothing
little smoothing
![Page 54: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/54.jpg)
Covariance Function § What is optimal in this case?
![Page 55: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/55.jpg)
Local Kernel Adaptation § Adapt kernels based on the terrain
gradients § Covariance is adjusted according to the
change in terrain elevation in the local neighborhood
local average
elevation gradient
![Page 56: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/56.jpg)
Adapting to Local Structures
Ground truth
Stationary GP Non-stationary GP
![Page 57: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/57.jpg)
Adapting to Local Structure § Model to deal with slowly changing
characteristics and strong discontinuities
![Page 58: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/58.jpg)
Experiments
standard
adaptive
![Page 59: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/59.jpg)
Experiments
Observation (with white noise σ=0.3)
Kernels Predicted Map Local errors
![Page 60: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/60.jpg)
Experiments – Stone Block
![Page 61: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/61.jpg)
Experiments – Stone Block
Ground Truth Observations
![Page 62: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/62.jpg)
Experiments – Stone Block
Prediction
![Page 63: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/63.jpg)
Experiments – Slope
Uncertainties:
Observations & model:
![Page 64: Advanced Techniques for Mobile Robotics Gaussian …ais.informatik.uni-freiburg.de/.../pdfs/rob2-14-gaussian-processes.pdf · Wolfram Burgard, Cyrill Stachniss, Kai Arras, Maren Bennewitz](https://reader030.vdocument.in/reader030/viewer/2022040623/5d4a8eca88c993d7648b8e32/html5/thumbnails/64.jpg)
Summary § GPs are a flexible and practical
approach to Bayesian regression § Prior knowledge is encoded in a
human understandable way § Learned models can be interpreted § Efficiency mainly depends on the
number of training points § Real-world problem sizes require
approximations/sparsity/…