Title: Mobile Robot Navigation in Outdoor Environments: A Topological Approach
1Mobile Robot Navigation in Outdoor
EnvironmentsA Topological Approach
Alberto Manuel Martinho Vale
Dissertação para a obtenção do Grau de Doutor
em Engenharia Electrotécnica e de Computadores
Instituto Superior Técnico Instituto de
Sistemas e Robótica
2Motivation - Novelties
- Topological maps, a type of representation with a
high level of abstraction - The three main problems (localization, navigation
and mapping) are addressed simultaneously - A linkage between this high level of abstraction
and motion commands - Dynamic Expectation-Maximization algorithm for
map building - The feature extraction and selection
methodologies - Probabilistic approach for localization and
navigation with a revisited Forward-Backward
algorithm
How did we come to PA3 ? Referential? Counting
the steps?
3Structure
- Mobile Robot Navigation
- Problem Definition
- Environment Representation
- The adopted representation Topological Maps
- Localization
- Markov Models (to support robot navigation)
- Changed version of Forward-Backward Algorithm
- Navigation
- Path Planning
- Low Level Driving Methodology
- Mapping
- Dynamic Expectation-Maximization Algorithm
- Features Extraction
- Features Selection
- Initialization
- Experimental Results
- Indoor
- Outdoor
- Conclusions and Future Directions
4The Mobile Robot Navigation Problem
Environment Model
5Environment Representation
6The adopted representation Topological Maps
- Provides crucial information to travel long
distances - Prepared to cope with the diversity of scenarios
(large spectrum of information) - Robust to scenario changes
- Useful for map sharing (mainly for heterogeneous
robots) - Improve hierarchical representation, i.e.,
different levels of topological representation
with different resolutions
7Structure
- Mobile Robot Navigation
- Problem Definition
- Environment Representation
- The adopted representation Topological Maps
- Localization
- Markov Models (to support robot navigation)
- Changed version of Forward-Backward Algorithm
- Navigation
- Path Planning
- Low Level Driving Methodology
- Mapping
- Dynamic Expectation-Maximization Algorithm
- Features Extraction
- Features Selection
- Initialization
- Experimental Results
- Indoor
- Outdoor
- Conclusions and Future Directions
8Localization
Given a sequence of observations, where is the
robot in the map?
9Log of Prob. Localization
The topological map is defined by 6 states, each
one characterized by 5 different type of features
10Structure
- Mobile Robot Navigation
- Problem Definition
- Environment Representation
- The adopted representation Topological Maps
- Localization
- Markov Models (to support robot navigation)
- Changed version of Forward-Backward Algorithm
- Navigation
- Path Planning
- Low Level Driving Methodology
- Mapping
- Dynamic Expectation-Maximization Algorithm
- Features Extraction
- Features Selection
- Initialization
- Experimental Results
- Indoor
- Outdoor
- Conclusions and Future Directions
11Navigation
12Low Level Navigation
- Reactive navigation (Behavior approach
Attractive and Repulsive behaviors) to transform
a sequence of states in a sequence of motion
commands
Motion Commands
Target direction
Attractive behavior
(the target direction is ?ij from state si to sj)
13Simulation Results Navigation
14Structure
- Mobile Robot Navigation
- Problem Definition
- Environment Representation
- The adopted representation Topological Maps
- Localization
- Markov Models (to support robot navigation)
- Changed version of Forward-Backward Algorithm
- Navigation
- Path Planning
- Low Level Driving Methodology
- Mapping
- Dynamic Expectation-Maximization Algorithm
- Features Extraction
- Features Selection
- Initialization
- Experimental Results
- Indoor
- Outdoor
- Conclusions and Future Directions
15Mapping Expectation-Maximization Algorithm
i 1 , 2 , ... , N
N states and the respective Gaussians aij ,
state transition probabilities ?ij , direction
between states
16Mapping Dynamic EM
Dynamic EM to determine N and based on the
entropy of state si
(wij probability of oi ? state sj)
The state transition probabilities aij and the
directions between states ?ij, are given
integrating the localization results in the
current map
17Topological Map Features
- Robustness to small displacements
- Invariant to lighting conditions
- Invariant to occlusion
- Fast computation
- Capacity to compress the images as much as
possible while retaining pertinent information
18Feature Selection
Select the best features to reduce the ambiguity
to the environment representation
and
19Features Correlation (results)
Exp.1 Pav. Central Exp.2
Garden Exp.3 Parking
Compression from observations to features
Correlation between features
20Summary
21Structure
- Mobile Robot Navigation
- Problem Definition
- Environment Representation
- The adopted representation Topological Maps
- Localization
- Markov Models (to support robot navigation)
- Changed version of Forward-Backward Algorithm
- Navigation
- Path Planning
- Low Level Driving Methodology
- Mapping
- Dynamic Expectation-Maximization Algorithm
- Features Extraction
- Features Selection
- Initialization
- Experimental Results
- Indoor
- Outdoor
- Conclusions and Future Directions
22Experimental Results (indoors)
Features acquired from range sensors average
free-area (measured by laser and by sonars) and
variance of free-area (by laser)
23Experimental Results (outdoors)
- Features acquired from images and orientation
sensors (colour and 2D histograms, edges and
orientation) - Results with PCA/ICA provide more ambiguities
Via points
24Experimental Results (outdoors)
Mapping
Localization
25Experimental Results (outdoors)
Mapping
Localization
26Structure
- Mobile Robot Navigation
- Problem Definition
- Environment Representation
- The adopted representation Topological Maps
- Localization
- Markov Models (to support robot navigation)
- Changed version of Forward-Backward Algorithm
- Navigation
- Path Planning
- Low Level Driving Methodology
- Mapping
- Dynamic Expectation-Maximization Algorithm
- Features Extraction
- Features Selection
- Initialization
- Experimental Results
- Indoor
- Outdoor
- Conclusions and Future Directions
27Conclusions
- It is possible to develop a topological approach
for the navigation at a high level of abstraction
and implemented in simple outdoor environments - Even when a sensor retrieves poor amount of
information, as a gyroscope - Localization, navigation and mapping are
addressed at the same level of abstraction - Scenario may be represented by a topological map
with a low resolution and therefore, with each
state incorporating another topological map or
even a metric map - Feature extraction and selection are important
issues of the topological approach, since the
representation is highly dependent to the type of
features
28Further Research
- The precision and resolution of a topological map
- Linkage between states (not necessarily by
orientations) - Combining the localization, navigation and
mapping together in a real situation of rescue
scenarios - Cooperative navigation develop the topological
approach using a team of robots - Different approaches to evaluate the features
correlation to improve the mapping algorithm - Features data processing (different types of
sensors)
29Mobile Robot Navigation in Outdoor
EnvironmentsA Topological Approach
Alberto Manuel Martinho Vale
Dissertação para a obtenção do Grau de Doutor
em Engenharia Electrotécnica e de Computadores
Instituto Superior Técnico Instituto de
Sistemas e Robótica