cos 495 - lecture 13 autonomous robot navigation · 2011. 11. 7. · 1 cos 495 - lecture 13...
TRANSCRIPT
![Page 1: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/1.jpg)
1
COS 495 - Lecture 13 Autonomous Robot Navigation
Instructor: Chris Clark Semester: Fall 2011
Figures courtesy of Siegwart & Nourbakhsh
![Page 2: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/2.jpg)
2
Control Structure
Perception
Localization Cognition
Motion Control
Prior Knowledge Operator Commands
![Page 3: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/3.jpg)
3
Localization: Outline
1. Localization Tools 1. Belief representation 2. Map representation 3. Probability Theory
2. Overview of Algorithms
![Page 4: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/4.jpg)
4
Belief Representation
§ Our belief representation refers to the method we describe our estimate of the robot state.
§ So far we have been using a Continuous Belief representation.
![Page 5: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/5.jpg)
5
Belief Representation
§ We can provide a description of the level of confidence we have in our estimate.
§ We typically use a Gaussian distribution to model the state of the robot.
§ We need to know the variance of this Gaussian!
![Page 6: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/6.jpg)
6
Belief Representation
§ For example, consider modeling our robot’s position with a 2D Gaussian:
![Page 7: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/7.jpg)
7
§ Continuous (single hypothesis)
§ Continuous (multiple hypothesis)
Belief Representation
![Page 8: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/8.jpg)
8
Belief Representation
§ Or, we could assign a probability of being in some discrete locations: Grid Topological
![Page 9: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/9.jpg)
9
Belief Representation
§ Discretized (prob. Distribution)
§ Discretized Topological (prob. dist.)
![Page 10: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/10.jpg)
10
Belief Representation
§ Continuous § Precision bound by sensor data § Typically single hypothesis pose estimate § Lost when diverging (for single hypothesis) § Compact representation § Reasonable in processing power
![Page 11: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/11.jpg)
11
Belief Representation
§ Discrete § Precision bound by resolution of discretization § Typically multiple hypothesis pose estimate § Never lost (when diverges converges to another cell). § Memory and processing power needed (unless
topological map used) § Aids discrete planner implementation
![Page 12: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/12.jpg)
12
Belief Representation
§ Multi-Hypothesis Example
![Page 13: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/13.jpg)
13
Localization: Outline
1. Localization Tools 1. Belief representation 2. Map representation 3. Probability Theory
2. Overview of Algorithms
![Page 14: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/14.jpg)
14
Map Representation
§ The application determines Map precision § The map and feature precisions must match the
sensor precisions § There is a clear trade-off between precision and
computational complexity § Similar to belief representations, there are two
main types: § Continuous § Discretized
![Page 15: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/15.jpg)
15
Map Representation
§ Continous line-based
![Page 16: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/16.jpg)
16
Map Representation
§ Exact cell decomposition
![Page 17: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/17.jpg)
17
Map Representation
§ Fixed cell decomposition
![Page 18: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/18.jpg)
18
Map Representation
§ Fixed cell decomposition
![Page 19: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/19.jpg)
19
Map Representation
§ Adaptive cell decomposition
![Page 20: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/20.jpg)
20
Map Representation
§ Topological decomposition
![Page 21: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/21.jpg)
21
Map Representation
§ Topological decomposition
![Page 22: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/22.jpg)
22
Map Representation
§ Topological decomposition
![Page 23: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/23.jpg)
23
Localization: Outline
1. Localization Tools 1. Belief representation 2. Map representation 3. Probability Theory
2. Overview of Algorithms
![Page 24: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/24.jpg)
24
Basic Probability Theory
§ Probability that A is true P(A) § We compute the probability of each robot state
given actions and measurements. § Conditional Probability that A is true given that
B is true P(A | B) § For example, the probability that the robot is at
position xt given the sensor input zt is P( xt | zt )
![Page 25: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/25.jpg)
25
Basic Probability Theory
§ Product Rule: p (A B ) = p (A | B ) p (B ) p (A B ) = p (B | A ) p (A ) § Can equate above expressions to derive Bayes rule.
§ Bayes Rule: p (A | B) = p (B | A ) p (A ) p (B )
![Page 26: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/26.jpg)
26
Localization: Outline
1. Localization Tools 2. Overview of Algorithms
1. Typical Methods 2. Basic Structure
![Page 27: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/27.jpg)
27
Localization Probabilistic Methods
§ Problem Statement: § Determine the state of a robot in a known
environment. § Strategy:
§ It might start to move from a known location, and keep track of its position using odometry.
§ However, the more it moves the greater the uncertainty in its position.
§ Therefore, it will update its position estimate using observation of its environment
![Page 28: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/28.jpg)
28
Localization Probabilistic Methods
§ Method: § Fuse the odometric position estimate with the
observation estimate to get best possible update of actual position
§
§ This can be implemented with two main functions: 1. Act 2. See
![Page 29: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/29.jpg)
29
Localization Probabilistic Methods
§ Action Update (Prediction) § Define function to predict position estimate based
on previous state xt-1 and encoder measurement ot or control inputs ut
x’t = Act (ot , xt-1)
§ Increases uncertainty
![Page 30: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/30.jpg)
30
Localization Probabilistic Methods
§ Perception Update (Correction) § Define function to correct position estimate x’t using
exteroceptive sensor inputs zt
xt = See (zt , x’t)
§ Decreases uncertainty
![Page 31: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/31.jpg)
31
Localization Probabilistic Methods
§ Motion generally improves the position estimate.
![Page 32: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/32.jpg)
32
Map-Based Localization Kalman Filtering vs. Markov
§ Markov Localization § Can localize from any
unknown position in map § Recovers from ambiguous
situation § However, to update the
probability of all positions within the whole state space requires discrete representation of space. This can require large amounts of memory and processing power.
§ Kalman Filter Localization § Tracks the robot and is
inherently precise and efficient
§ However, if uncertainty grows too large, the KF will fail and the robot will get lost.
![Page 33: COS 495 - Lecture 13 Autonomous Robot Navigation · 2011. 11. 7. · 1 COS 495 - Lecture 13 Autonomous Robot Navigation Instructor: Chris Clark Semester: Fall 2011 Figures courtesy](https://reader035.vdocuments.us/reader035/viewer/2022071517/613ac865f8f21c0c8268a1eb/html5/thumbnails/33.jpg)
33
Particle Filter Example