Lecture 7 Localization

8/14/2019 Lecture 7 Localization

1/39

CpE 521A

A Introduction to Autonomous Mobile Robots

Lecture 7: Localization and Map Building

Part 1

Yan Meng

Department of Electrical and Computer Engineering

Stevens Institute of Technology


2/39

Localization and Map Building

Noise and aliasing; odometric position estimation

To localize or not to localize

Belief representation Map representation

"Position"Global Map

Perception Motion Control

Cognition

Real WorldEnvironment

Localization

PathEnvironment ModelLocal Map


3/39

Localization, Where am I?

Odometry, Dead Reckoning

Localization base on external sensors,

beacons or landmarks

Probabilistic Map Based Localization

Perception


4/39

Challenges of Localization

GPS may be the answer?

Knowing the absolute position (e.g. GPS) is not sufficient

GPS can not function indoors or in obstructed areas

Localization in human-scale in relation with environment

Planning in the Cognition step requires more than only position as input

It may need to acquire or build a map

Localization actually means building a map, then identifying the robotsposition relative to the map

Perception and motion play important roles

Sensor noise

Sensor aliasing

Effector noise

The inaccuracy and incompleteness of sensors and effectors pose thedifficult challenges to localization


5/39

Sensor Noise

Sensor noise induces a limitation on the consistency of sensor readingsin the same environmental state

Sensor noise is mainly influenced by environmente.g. surface, illumination

or by the measurement principle itselfe.g. interference between ultrasonic sensors

Sensor noise drastically reduces the useful information of sensorreadings. The solution is:

to take multiple reading into account

employ temporal and/or multi-sensor fusion


6/39

Sensor Aliasing

Unlike human sensing system, in robots, non-uniqueness of sensors

readings is the norm

Example: sonar or laser rangefinder

Even with multiple sensors, there is a many-to-one mapping from

environmental states to robots perceptual inputs

Therefore the amount of information perceived by the sensors is

generally insufficient to identify the robots position from a singlereading

Robots localization is usually based on a series of readings

Sufficient information is recovered by the robot over time


7/39

Effector Noise: Odometry and 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/orheading sensors is integrated to the position.

Pros: Straight forward, easy

Cons: Errors are integrated -> unbound

Using additional heading sensors (e.g. gyroscope) might help to reduce

the cumulated errors, but the main problems remain the same.


8/39

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 alwaysleading 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, not planar )


9/39

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

robots angular orientationOver long periods of time, turn and drift errors

far outweigh range errors!

Consider moving forward on a straight line along the x axis. The errorin the y-position introduced by a move ofdmeters will have a component

ofdsin, which can be quite large as the angular error grows.


10/39

Odometry: The Differential Drive Robot (1)

Position can be estimated starting from a know position p by

integrating the movement

= y

x

p

+= y

x

pp


11/39


Kinematics


12/39


Error model


13/39

Odometry: Growth of Pose Uncertainty for Straight Line Movement

Note: Errors perpendicular to the direction of movement are growing much faster!


14/39

Odometry: Growth of Pose uncertainty for Movement on a Circle

Note: Errors ellipse in does not remain perpendicular to the direction of movement!


15/39

Odometry: Calibration of Errors I (Borenstein [5])

The unidirectional square path experiment

BILD 1 Borenstein


16/39

Odometry: Calibration of Errors II (Borenstein [5])

The bi-directional square path experiment

BILD 2/3 Borenstein


17/39

To localize or not?

How to navigate between A and B

navigation without hitting obstacles

detection of goal location

Possible by always following the left wall

However, how to detect that the goal is reached


18/39

Behavior Based Navigation

Since sensors and effectors are noisy and information-limited, one may

want to design sets of behaviors instead of creating a geometric map

for localization.

Assume there exists a procedural solution to the particular navigation

problem at hand.


19/39

Behavior Based Navigation

Advantage

May be implemented very quickly for a single environment with a small

number of goal positions

Disadvantages

Does not directly scale to other environments or to larger environments The underlying procedures must be carefully designed to produce the

desired behavior ( time-consuming and environmental-dependent)

The fusion and rapid switching between multiple behaviors can negate

the fine-tuning procedure, and the addition of new behavior forces the

designer to retune all of the existing behaviors again


20/39

Model Based Navigation

Explicitly attempts to localize by collecting sensor data, then updating somebelief about its position with respect to a map of the environment


21/39


22/39

Belief Representation

a) Continuous map

with single hypothesis

b) Continuous mapwith multiple hypothesis

d) Discretized map

with probability distribution

d) Discretized topological

map with probability

distribution


23/39

Belief Representation: Characteristics

Continuous

Precision bound by sensor

dataTypically single hypothesis

pose estimate

Lost when diverging (forsingle hypothesis)

Compact representation and

typically reasonable inprocessing power.

Discrete

Precision bound by

resolution of discretisationTypically multiple hypothesis

pose estimate

Never lost (when divergesconverges to another cell)

Important memory and

processing power needed.(not the case for topological

maps)


24/39

Single-hypothesis Belief Continuous Maps

Real map with walls, doors, and furniture Line-based map


25/39

Single-hypothesis Belief Grid and Topological Map

Occupancy grid-based mapTopological map using

line features and doors


26/39

Single Hypothesis Belief

Advantage:

No position ambiguity

Make the decision-making much easier

Disadvantage:

Always generate a single hypothesis for position update is challengingdue to the effector and sensor noise


27/39

Multiple-hypothesis Belief

The robot tracks not just a single possible position but a possibly

infinite set of positions

One way to represent the set of possible robot positions is to usemultiple Gaussian probability density functions

Advantages

Maintain a sense of position while explicitly annotating the robotsuncertainty about its own position

Enable robots with limited sensory information to navigate robustly

Disadvantages

Make the decision-making more difficult

Some of the robots possible positions imply a motion trajectory that is

inconsistent with some of its other possible positions

Computational expensive


28/39

Grid-base Representation - Multi Hypothesis

Grid size around 20 cm2.

Courtesy of W. Burgard


29/39

Map Representation

1. Map precision vs. application

2. Features precision vs. map precision

3. Precision vs. computational complexity

Continuous Representation

Decomposition (Discretization)


30/39

Representation of the Environment

Environment Representation

Continuos Metric x,y,

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 valueso makes use of all acquired information

Low level features, e.g. line other geometric featureso 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


31/39

Map Representation: Continuous Line-Based

a) Architecture map

b) Representation with set of infinite lines


32/39

Map Representation: Decomposition (1)

Exact cell decomposition

Exact decomposition is not always feasible in real-world


33/39


Fixed cell decomposition

Obstacle-filled or free area

Narrow passages disappear


34/39


Adaptive cell decomposition


35/39


Occupancy grid with very small cells

Most common map representation technique currently utilized

Memory size may become untenable with large size of environment, not

compatible with closed-world assumption

Courtesy of S. Thrun


36/39


Topological Decomposition


37/39



node

Connectivity(arcs)


38/39



State-of-the-Art: Current Challenges in Map


39/39

Representation

Real world is dynamic

Differentiate permanent obstacles (e.g., walls, doorways, etc.) andtransient obstacles (e.g., humans, shipping packages, etc.)

Perception is still a major challenge (error prone, extraction of usefulinformation difficult)

Traversal of open space

Traditional range sensors are difficult for wide-open spaces, such asparking lots, fields of grass, and indoor atriums, because of their relativesparseness

How to build up topology (boundaries of nodes) in wide-open area?

GPS may be one solution

Sensor fusion

A variety of sensor types can have their data correlated appropriately,

obtain the perceptions well beyond that of any individual one More

Lecture 7 Localization

Documents

Transcript of Lecture 7 Localization