ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
JESUS TORDESILLAS TORRES, 2019
2
3
4
Traversability
5
6
7
8
PaperMethod
Map used + Handle Errors in Estimation?
Planning
9
Bayesian Generalized Kernel Inference for Terrain Traversability Mapping
Bayesian Inference:
Assume Perfect EstimationD* + Controller to track it
10
11
12
1.-The height map is obtained through a Bay. Kernel Regression ​
13
2.-The traversability map is obtained with a weighted sum of step height, slope and roughness​
14
3.-Using a Bay. Kernel Classification, cells are classified as traversable or not as a function of the traversability value and varianc
15
Stereo camera based navigation of mobile robots on rough terrain
Traversability is a weighted sum of:
Simply merge maps, and play with the max/min height thresholds to "filter" the errors in the map due to estimation errorsNo
16
•Slope: Fit Plane + Normal Estimation
17
•Roughness: Std Deviation of the height values
18
•Step Height: Max height difference using a square window patch
19
Driving on Point Clouds
Traversability is a function of:
ICP to estimate transformation. Use also a stack of old maps that are used in the case when ICP doesn't work with the two most recents point cloudsTrajectory = spline of 3rd degree polynomials, each of one is on a planar surface. These splines are optimized and locally modified to take into account traversability, nonholonomic and curvature constraints
20
•Roll and pitch of the vehicle in that position
21
•Roughness (residual fitting a plane)
22
23
24
Learning Traversability from Point Clouds in Challenging ScenariosKernel SVN, with feature vectors based on:Use instantaneous point cloudNo
25
•Geometry:
26
 •Covariance Features: Linearity, Planarity, Sphericity, Omnivariance ,…
27
 •Roughness-Based Features: Slope, Goodness of fit, height,…
28
 •Others: Normal Vector, Unevenness, Surface Density,...
29
•Appearance: c1c2c3 color model, augmented with hue
30
Terrain traversability analysis methods for unmaned ground vehicles: A survey
The traversability approaches are classified into:
31
--> Proprioceptive: Analyze vibrations, wheel slips, bumper hits,.. when the robot is traversing the terrain



32
--> Geometry-Based: They take into account the terrain model, stability constraints, kinematic constraints and/or the robot model
33
34
35
36
37
38
39
40
41
42
43
44
45
--> Appearance Based: Segment the different terrains in the image plane or in the point cloud. Many learning-based methods here
46
Probabilistic Terrain Analysis For High-Speed Desert Driving
47
(I'm reading it right now). They basically take into account the estimation error and the sensor noise and use an MDP. The parameters of the MPD are trained using labeled data.
48
49
50
51
52
MAPPING
53
54
55
PaperMethod
56
57
Voxblox: Incremental 3D Euclidean Signed Distance Fields for On-Board MAV PlanningBuild ESDF (Euclidean Signed Distance Fields) from TSDF (Truncated Signed Distance Fields). It also generates a mesh
58
Can incorporate uncertainty in the pose into the merging process?
59
NanoMap: Fast, Uncertainty-Aware Proximity Queries with Lazy Search over Local 3D DataKeep a stack of all the last N point clouds, with an uncertainty associated with each one and with the transformation between each pair of them
60
When a trajectory is generated, the collision check is performed using the point cloud that contains the bounding box of the 1-sigma volume of the distribution of the query point.
61
No map is built
62
63
Aggressive 3-D collision avoidance for high-speed navigation(from Brett, for UAVs)Use only instateneous Point Cloud (from a RS200), the trajectory is forced to be inside the field of view
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100