mobile robot localization techniques and methods

Autonomous Mobile Robots, Chapter 5
© R. Siegwart, I. Nourbakhsh
Localization, Where am I?
?
• Odometry, Dead Reckoning
• Localization base on external sensors,
beacons or landmarks
• Probabilistic Map Based Localization
5.1
Observation
Map
data base
Prediction of
Position
(e.g. odometry)
Perception
Matching
Position Update
(Estimation?)
raw sensor data or
extracted features
predicted position
position
matched
observations
YES
Encoder
Perception

Challenges of Localization
• Knowing the absolute position (e.g. GPS) is not sufficient
• Localization in human-scale in relation with environment
• Planning in the Cognition step requires more than only position as input
• Perception and motion plays an important role
 Sensor noise
 Sensor aliasing
 Effector noise
 Odometric position estimation
5.2

Sensor Noise
• Sensor noise is mainly influenced by environment
e.g. surface, illumination …
• or by the measurement principle itself
e.g. interference between ultrasonic sensors
• Sensor noise drastically reduces the useful information of sensor
readings. The solution is:
 to take multiple reading into account
 employ temporal and/or multi-sensor fusion
5.2.1

Sensor Aliasing
• In robots, non-uniqueness of sensors readings is the norm
• Even with multiple sensors, there is a many-to-one mapping from
environmental states to robot’s perceptual inputs
• Therefore the amount of information perceived by the sensors is
generally insufficient to identify the robot’s position from a single
reading
 Robot’s localization is usually based on a series of readings
 Sufficient information is recovered by the robot over time
5.2.2

Effector Noise: Odometry, Dead Reckoning
• Odometry and dead reckoning:
Position update is based on proprioceptive sensors
 Odometry: wheel sensors only
 Dead reckoning: also heading sensors
• The movement of the robot, sensed with wheel encoders and/or
heading sensors is integrated to the position.
 Pros: Straight forward, easy
 Cons: Errors are integrated -> unbounded
• Using additional heading sensors (e.g. gyroscope) might help to reduce
the accumulated errors, but the main problems remain the same.
5.2.3

Odometry: Error sources
deterministic non-deterministic
(systematic) (non-systematic)
 deterministic errors can be eliminated by proper calibration of the system.
 non-deterministic errors have to be described by error models and will always
lead to uncertain position estimate.
• Major Error Sources:
 Limited resolution during integration (time increments, measurement resolution
…)
 Misalignment of the wheels (deterministic)
 Unequal wheel diameter (deterministic)
 Variation in the contact point of the wheel
 Unequal floor contact (slipping, non-planar …)
 …
5.2.3

Odometry: Classification of Integration Errors
• Range error: integrated path length (distance) of the robots movement
 sum of the wheel movements
• Turn error: similar to range error, but for turns
 difference of the wheel motions
• Drift error: difference in the error of the wheels leads to an error in the
robot’s angular orientation
Over long periods of time, turn and drift errors
far outweigh range errors!
 Consider moving forward on a straight line along the x axis. The error
in the y-position introduced by a move of d meters will have a component
of dsin∆θ, which can be quite large as the angular error ∆θ grows.
5.2.3

Odometry: The Differential Drive Robot (1)










θ
= y
x
p










θ
∆
∆
∆
+
=
′ y
x
p
p
5.2.4

• Kinematics
5.2.4

• Error model (details are beyond the scope of our class; just know that
we can build an error model…)
5.2.4

Odometry: Growth of Pose uncertainty for Straight Line Movement
• Note: Errors perpendicular to the direction of movement are growing much faster!
5.2.4
(ellipses
represent
uncertainty in
position)

Odometry: Growth of Pose uncertainty for Movement on a Circle
• Note: Error ellipse does not remain perpendicular to the direction of movement!
5.2.4

To localize or not?
• How to navigate between A and B
 navigation without hitting obstacles
 detection of goal location
• Possible by following always the left wall
 However, how to detect that the goal is reached
5.3

Belief Representation
• a) Continuous map
with single hypothesis
• b) Continuous map
with multiple hypotheses
• d) Discretized map
with probability distribution
• d) Discretized topological
map with probability
distribution
5.4

Belief Representation: Characteristics
• Continuous
 Precision bound by sensor
data
 Typically single hypothesis
pose estimate
 Lost when diverging (for
single hypothesis)
 Compact representation and
typically reasonable in
processing power.
• Discrete
 Precision bound by
resolution of discretization
 Typically multiple hypothesis
pose estimate
 Never lost (when diverges
from one cell, it converges to
another cell)
 Important memory and
processing power needed.
(not the case for topological
maps)
5.4

Single-hypothesis Belief – Continuous Line-Map
5.4.1

Single-hypothesis Belief – Grid and Topological Map
5.4.1

Grid-based Representation – Multi-Hypothesis
• Grid size around 20 cm2.
• Clouds represent possible robot locations
• Darker coloring means higher probability
5.4.2
Courtesy of W. Burgard

Map Representation
1. Map precision vs. application
• The precision of the map must match the precision with which the robot needs
to achieve its goals.
2. Features precision vs. map precision
• The precision of the map and the type of features represented must match the
precision and data types returned by the robot’s sensors.
3. Precision vs. computational complexity
• The complexity of the map representation has a direct impact on the
computational complexity of reasoning about mapping, localization, and
navigation
Two primary map choices:
• Continuous Representation
• Decomposition (Discretization)
5.5

Representation of the Environment
• Environment Representation
 Continuous Metric → x,y,θ
 Discrete Metric → metric grid
 Discrete Topological → topological grid
• Environment Modeling
 Raw sensor data, e.g. laser range data, grayscale images
o large volume of data, low distinctiveness on the level of individual values
o makes use of all acquired information
 Low level features, e.g. line other geometric features
o medium volume of data, average distinctiveness
o filters out the useful information, still ambiguities
 High level features, e.g. doors, a car, the Eiffel tower
o low volume of data, high distinctiveness
o filters out the useful information, few/no ambiguities, not enough information
5.5

Map Representation: Continuous Line-Based
a) Architecture map
b) Representation with set of infinite lines
5.5.1

Map Representation: Decomposition (1)
• Exact cell decomposition
5.5.2

• Fixed cell decomposition
 Narrow passages can disappear
5.5.2

• Adaptive cell decomposition (i.e., quadtree)
5.5.2

• Fixed cell decomposition – Example with very small cells
5.5.2
Courtesy of S. Thrun

• Topological Decomposition
5.5.2

node
Connectivity
(arch)
5.5.2

~ 400 m
~ 1 km
~ 200 m
~ 50 m
~ 10 m
5.5.2

State-of-the-Art: Current Challenges in Map Representation
• Real world is dynamic
• Perception is still a major challenge
 Error prone
 Extraction of useful information difficult
• Traversal of open space
• How to build up topology (boundaries of nodes)
• Sensor fusion
• …
5.5.3

Probabilistic, Map-Based Localization (1)
• Consider a mobile robot moving in a known environment.
• As it starts to move, say from a precisely known location, it might
keep track of its location using odometry.
• However, after a certain movement the robot will get very uncertain
about its position.
 update using an observation of its environment.
• Odometric information leads to an estimate of the robot’s position,
which can then be fused with the sensor observations to get the best
possible update of the robot’s actual position.
5.6.1

Probabilistic, Map-Based Localization (2)
• Action update:
 action model Act:
where ot : Encoder Measurement, st-1 : prior belief state
 increases uncertainty
• Perception update:
 perception model See:
where it : exteroceptive sensor inputs, s′t : updated belief state
 decreases uncertainty
5.6.1
1
( , )
t t t
s Act o s −
′ =
( , )
t t t
s See i s′
=

The Five Steps for Map-Based Localization
Observation
on-board sensors
Map
data base
Prediction of
Measurement and
Position(odometry)
P
e
r
c
e
p
t
i
o
n
Matching
Estimation
(fusion)
raw sensor data or
extracted features
p
r
e
d
i
c
t
e
d
f
e
a
t
u
r
e
o
b
s
e
r
v
a
t
i
o
n
s
position
estimate
matched predictions
and observations
YES
Encoder
1. Prediction based on previous estimate and odometry
2. Measurement prediction based on prediction and map
3. Observation with on-board sensors
4. Matching of observation and map
5. Estimation -> position update (posteriori position)
5.6.1

Two general approaches:
Markov and Kalman Filter Localization
• Markov localization
 Maintains multiple estimates of
robot position
 Localization can start from any
unknown position
 Can recover from ambiguous
situations
 However, to update the probability
of all positions within the state
space requires a discrete
representation of the space (grid);
if a fine grid is used (or many
estimates are maintained), the
computational and memory
requirements can be large.
• Kalman filter localization
 Single estimate of robot position
 Requires known starting position
of robot
 Tracks the robot and can be very
precise and efficient
 However, if the uncertainty of the
robot becomes too large (e.g. due
collision with an object) the
Kalman filter will fail and the
robot becomes “lost”.
5.6.1

Three types of localization problems
• “Global” localization – figure out where the robot is, but we don’t know
where the robot started
• “Position tracking” – figure out where the robot is, given that we know
where the robot started
• “Kidnapped robot” – robot is moved by external agent
The Markov / Monte Carlo Localization approach of Fox, et al (which is
in Player/Stage) can address all 3 problems

Markov Localization
• Markov localization uses an explicit, discrete representation for the
probability of all positions in the state space.
 Later, we’ll talk about a more efficient version (called Monte Carlo
localization) that randomly samples possible positions, instead of
maintaining information about all positions
• This is usually done by representing the environment by a grid or a
topological graph with a finite number of possible states (positions).
• During each update, the probability for each state (element) of the
entire space is updated.
5.6.2

Markov Localization
• Key idea: compute a probability distribution over all possible positions
in the environment.
 This probability distribution represents the likelihood that the robot is in a
particular location.
P(Robot Location)
X
Y
State space = 2D, infinite #states
Slide adapted from Dellaert presentation “19-Particles.ppt”

Markov Localization makes use of Bayes Rule
• P(A): Probability that A is true.
 e.g. p(rt = l): probability that the robot r is at position l at time t
• We wish to compute the probability of each individual robot position
given actions and sensor measures.
• P(A|B): Conditional probability of A given that we know B.
 e.g. p(rt = l| it): probability that the robot is at position l given the
sensors input it.
• Product rule:
• Bayes rule:
5.6.2

The “See” update step
• Bayes rule:
 “See” operation: Maps from a belief state and a sensor input to a refined
belief state:
(5.21)
 p(l): belief state before perceptual update process
 p(i |l): probability we get measurement i when being at position l
o To obtain this info: consult robot’s map and identify the probability of a certain sensor
reading if the robot were at position l
 p(i): normalization factor so that sum over all l equals 1.
• We apply this operation to all possible robot positions, l
5.6.2
( , )
t t t
s See i s′
=

The “Act” update step
 “Act” operation: Maps from a belief state (i.e., belief in robot being in
some prior position) and an action (represented by ot, which is the encoder
measurement corresponding to an action) to a new belief state:
(5.22)
 This operation sums over all possible ways in which the robot may have
reached position l at time t, from any possible prior position at time t-1.
(Note that there can be more than 1 way to reach a given position, due to
uncertainty in encoder measurement.)
5.6.2
1
( , )
t t t
s Act o s −
′ =
1 1 1
( | ) ( | , ) ( )
t t t t t t t
p l o p l l o p l dl
− − −
′ ′ ′
= ∫

The Markov Property
• These two updates (from prior 2 slides):
“See”: (5.21)
“Act”: (5.22)
constitute the Markov assumption. That is, the current update only depends
on the previous state (lt) and its most recent action (ot) and perception (it).
The Markov assumption may not be true, but it greatly simplifies tracking,
reasoning, and planning, so it is a common approximation in robotics/AI.
5.6.2
1 1 1
( | ) ( | , ) ( )
t t t t t t t
p l o p l l o p l dl
− − −
′ ′ ′
= ∫

Markov Localization: Case Study – Grid Map
• Fine fixed decomposition grid (x, y, θ), 15 cm x 15 cm x 1°
• Action update:
 Sum over previous possible positions
and motion model:
 (this is discrete version of eqn. 5.22)
• Perception update:
 Given perception i, what is the
probability of being in location l:
5.6.2
Courtesy of
W. Burgard

Perception update details
• The critical challenge is the calculation of p(i|l)
 p(i|l) is computed using a model of the robot’s sensor behavior, its position l, and the
local environment metric map around l.
 Assumptions:
o Measurement error can be described by a distribution with a mean at the correct reading
o Non-zero chance for any measurement
o Local peak at maximal reading of range sensor (due to absorption/reflection)
5.6.2
Courtesy of
W. Burgard

Markov Localization: General idea for maintaining
multiple estimates of robot position
• The 1D case
1. Start
No knowledge at start, thus we have
a uniform probability distribution.
2. Robot perceives first pillar
Seeing only one pillar, the probability
being at pillar 1, 2 or 3 is equal.
3. Robot moves
Action model enables estimation of the
new probability distribution based
on the previous one and the motion.
4. Robot perceives second pillar
Based on all prior knowledge, the
probability of being at pillar 2
becomes dominant
5.6.2

Markov Localization: Example
• Example 1: Office Building
5.6.2
Position 3
Position 4
Position 5
Courtesy of
W. Burgard

Markov Localization: Example #2
• Example 2: Museum
 Laser scan 1
5.6.2
Courtesy of
W. Burgard

 Laser scan 2
5.6.2
Courtesy of
W. Burgard

 Laser scan 3
5.6.2
Courtesy of
W. Burgard

 Laser scan 13
5.6.2
Courtesy of
W. Burgard

 Laser scan 21
5.6.2
Courtesy of
W. Burgard

Markov Localization: Using Randomized Sampling to
Reduce Complexity  “Particle filters”, “Monte Carlo algorithms”
• Fine fixed decomposition grid results in a huge state space
• Reducing complexity:
 The main goal is to reduce the number of states that are updated in each
step
• Randomized Sampling / Particle Filters
 Approximated belief state by representing only a ‘representative’ subset
of all states (possible locations)
 E.g., update only 10% of all possible locations
 The sampling process is typically weighted, e.g., put more samples
around the local peaks in the probability density function
 However, you have to ensure some less likely locations are still tracked,
otherwise the robot might get lost
5.6.2

Updating beliefs using Monte Carlo Localization (MCL)
• As before, 2 models: Action Model, Perception Model
• Robot Action Model:
 When robot moves, MCL generates N new samples that approximate robot’s
position after motion command.
 Each sample is generated by randomly drawing from previous sample set, with
likelihood determined by p values.
 For sample drawn with position l′, new sample l is generated from P(l | l′, a), for
action a
 p value of new sample is 1/N
(From Fox, et al, AAAI-99)
Sampling-based
approximation
of position belief for
non-sensing robot

• Robot Perception Model:
 Re-weight sample set according to the likelihood that robot’s current
sensors match what would be seen at a given location
• After applying Motion model and Perception model:
 Resample, according to latest weights
 Add a few uniformly distributed, random samples
o Very helpful in case robot completely loses track of its location
 Go to next iteration
Updating beliefs using Monte Carlo Localization (MCL), con’t.

Example Results
Initially, robot doesn’t know where it is
(see particles representing possible robot locations
distributed throughout the environment)
After robot moves some, it gets better estimate
(see particles clustered an a few areas, with a
few random particles also distributed around for
robustness)

Adapting the Size of the Sample Set
• Number of samples needed to achieve a desired level of accuracy varies
dramatically depending on the situation
 During global localization: robot is ignorant of where it is  need lots of
samples
 During position tracking: robot’s uncertainty is small  don’t need as
many samples
• MCL determines sample size “on the fly”
 Compare P(l) and P(l | s) (i.e., belief before and after sensing) to
determine sample size
 The more divergence, the more samples that are kept

Player/Stage Localization Approach:
Monte Carlo Localization
• Based on techniques developed by Fox, Burgard, Dellaert, Thrun
(AAAI’99)
(Movie illustrating approach)

More movies
• Dieter Fox movie: MCL using Sonar
• Dieter Fox movie: MCL using Laser

Summarizing the process: Particle Filtering
weighted S′
t St
S′
t
St-1
Predict ReWeight Resample
Slide adapted from Dellaert presentation “19-Particles.ppt”

What sensor to use for localization?
• Can work with:
 Sonar
 Laser
 Vision
 Radio signal strength

mobile robot localization techniques and methods

More Related Content

Similar to mobile robot localization techniques and methods (20)

More from Jenitha Rajadurai (20)

Recently uploaded (20)

mobile robot localization techniques and methods