CSE 461

Introduction to Robotics
Navigation : Path planning, Localization, Mapping
,Exploration
Abdulla Hil Kafi
Research Associate,
School of Engineering,
Brac University
4/28/2021 CSE461 : Introduction to Robotics 1
Navigation
• The act, activity, or process of finding the way to get to a
place
• Navigation is the ability to determine your location and plan
a path to some goal

4/28/2021 CSE461 : Introduction to Robotics 2

Robot Navigation
For autonomous behavior, robots need the ability to navigate:
• Learn the environment->“Model”
• Estimate where it is in the environment->“Localize”
• Move to desired locations in the environment

4/28/2021 CSE461 : Introduction to Robotics 3

Navigation Example
GOOGLE CAR DILIGENT
Scenarios
•Hospital Helper
(e.g. Diligent, Tugs)
•Office security or mail-
delivery
(e.g.Cobal, Savioke)
•Tour Guide robot in a
COBALT(office) museum (Minerva)
SAVIOKE •Autonomous Car with GPS
(hotels)
and Nav system
Biological analogies:
Humans, bees and ants,
migrating birds, herds

4/28/2021 CSE461 : Introduction to Robotics 4

Navigation Problem
Problem Characteristics
• Environments are Known versus Unknown
• Environments are Static versus Dynamic
• Environments are Structured versus Unstructured (Indoors
versus Outdoors?)

4/28/2021 CSE461 : Introduction to Robotics 5

Robots Navigating
•Path Planning: How I get to my Goal?
•Localization: Where am I?
•Mapping: Where have I been?
•Exploration: Where haven’t I been?

4/28/2021 CSE461 : Introduction to Robotics 6

What is Path Planning?
• Simple Question: How do I get to my Goal?
• Not a simple answer!
• Can you see your goal?
• Do you have a map?
• Are obstacles unknown or dynamic?
• Does it matter how fast you get there?
• Does it matter how smooth the path is ?
• How much compute power do you have?
• How precise is your motion control?
• Path Planning is best thought of as a Collection of Algorithms
• 3 Things need to consider: Environment, Success metrics, Robot
capability.
4/28/2021 CSE461 : Introduction to Robotics 7
Types of Path Planning Approaches
• Basics
• Visual homing (Purely local sensing and feedback control)
• Inverse Kinematics (Turn-move-turn to get from A to B)
• Bug-based Path Planning (mostly-local without a map)
• Robots can see the Goal (direction and distance)
• But there are unknown obstacles in the way (No map)
• Metric (A*) Path Planning (global with a map)
• Assumes that you have a map (distance or graph)and you know
where you and the goal are located in it.
• Path is represented as a of series of waypoints(directions)

4/28/2021 CSE461 : Introduction to Robotics 8

Basics: Visual Homing
Purely Reactive Navigation
• Measure Visual (x,y) Position of Goal
• Move to bring goal to Visual Center
• Proportional Control (if you see the goal), Random walk (if
you don’t)

4/28/2021 CSE461 : Introduction to Robotics 9

Bug-based Path Planning
➢What if the Robot has obstacles in the way?
➢Always have Goal direction and/or distance (Global)
➢But No Map: Only local knowledge of environment (Local)
➢Example Scenario:
Outdoor robot knows GPS location of goal, but building in the way.
Indoor robot see goal location, but furniture in the way.

➢ “Bug” Algorithms depend on simple but provable

behaviors!
➢ Don’t need to build a map
➢ Simple Computation: Visual Homing + Wall-following + Odometry

➢Very intuitive class of algorithms – but surprisingly powerful

4/28/2021 CSE461 : Introduction to Robotics 10

Metric/Global Path Planning
➢What if the Robot has Full Knowledge
➢A map of the environment and robot + goal’s locations
➢Goal: Find a “optimal” path (typically distance)

➢Two Components
➢Map Representation (“graph”):
➢ Feature based maps (office numbers, landmarks)
➢ Grid based maps (cartesian, quadtrees)
➢ Polygonal maps (geometric decompositions)
➢Path Finding Algorithms:
➢ Shortest-Path Graph Algorithms (Breadth-First-Search, A* Algorithm)

4/28/2021 CSE461 : Introduction to Robotics 11

Map Representation: Feature based
➢ Also known as a Topological or Landmark-based Map
 Features your robot can recognize:
 Includes both natural landmarks (corner, doorway, hallways) and artificial ones (office
door numbers; or robot-friendly tags)
 Gateways are landmarks that represent decisions (e.g. intersection)
 Distinguishable places are unique landmarks
➢ World is a graph that connects landmarks
 Edges represent actual motion: how to get from landmark A to landmark B
Usually visual/reactive navigation is possible along an edge
 Edges can also keep extra attributes: distance, time it takes, etc.
➢ Google Maps are topological maps for humans (e.g. turn at intersection)

4/28/2021 CSE461 : Introduction to Robotics 12

Path Finding Algorithms
➢All Map Representations are a weighted “graph”
➢Nice part is that you only need to do this once
➢Algorithm: Compute shortest paths in the graph
➢Path is represented by a series of waypoints
➢Single Path Search Algorithms: Find shortest path A to B
➢Breadth-First-Search (simple graphs);
➢A* search for large graphs (BFS + Heuristic)
➢Gradient Path Algorithms: Find all paths towards B
➢E.g. Fixed Base station: BFS, Dijkstra’s, Wavefront algorithms, etc.

4/28/2021 CSE461 : Introduction to Robotics 13

Robots Navigating
•Path Planning: How to I get to my Goal?
•Localization: Where am I?
•Mapping: Where have I been?
•Exploration: Where haven’t I been?

4/28/2021 CSE461 : Introduction to Robotics 14

Localization
➢Simple Question: Where am I?
➢Not a simple answer:
➢Do you have a map?
Yes => a global position in the world
No => position in reference to other objects? Or your own past?
➢What can you sense?
Can you sense and record your own self-movement?
Can you sense external things like landmarks?
How certain are you about what you sense?
➢Localization is a “collection of algorithms”
4/28/2021 CSE461 : Introduction to Robotics 15
Localization Techniques
➢Dead-reckoning (motion)
➢ Keep track of where you are without a map, by recording the series of actions that you made,
using internal sensors. (also called Odometry, Path Integration)

➢Landmarks (sensing)
➢ Triangulate your position geometrically, by measuring distance to one or more known
landmarks
E.g. Visual beacons or features, Radio/Cell towers and signal strength, GPS!

➢State Estimation (uncertainty in motion & sensing)

➢ Probabilistic Reasoning
➢ Kalman Filters (combine both motion and sensing)
➢ Particle Filters (also known as Monte Carlo Localization)

4/28/2021 CSE461 : Introduction to Robotics 16

Dead-Reckoning
• Keep track of initial position and the series of
movements/actions that you made.
• Method: Take a “step”, compute new position.
• Also called odometry or path integration
Example: Inertial navigation systems (INS)
• Complex motion (momentum, external effects)
• Include accelerometers and gyroscopes to provide better measurements of instantaneous
velocity.
• Expensive systems very good
• satellites, submarines
• But, low-cost IMU’s increasingly available

4/28/2021 CSE461 : Introduction to Robotics 17

Land Mark
• How it works
 Opposite of dead-reckoning!
 Use measurements to external landmarks of known position
 Examples: visual landmarks, radio towers, GPS!

• Example
•  GPS Satellites are your “landmarks”
 Continually transmits a message
 Message includes both time of
transmission, and satellite position
 GPS Receiver
 Compute distance by measuring signal
transmission time (speed of light)
 3D: Lie on the intersection of 4 spheres!

4/28/2021 CSE461 : Introduction to Robotics 18

State Estimation (uncertainty in motion & sensing)
• Key Idea: Combine Motion and Sensing
• (Dead-reckoning + uncertainty) + (Landmarks + uncertainty)
• Each has error, but the error can be complementary
• Kalman Filters
• Take advantage of mathematics of Gaussians to model uncertainty
• General method for state estimation (not just localization)
• Applications: Car + GPS, Lawnmower + beacons, warehouse robots
• Particle Filters (Monte Carlo Localization)
• Use a discrete distribution of “Particles” to represent uncertainty
think of sampling or histograms)
• Useful when environment is complex and ambiguous
• Application: A robot wandering in a building with a map

4/28/2021 CSE461 : Introduction to Robotics 19

Robots Navigating
•Path Planning: How to I get to my Goal?
•Localization: Where am I?
•Mapping: Where have I been?
•Exploration: Where haven’t I been?

4/28/2021 CSE461 : Introduction to Robotics 20

Mapping and Exploration
• Question:
• You are roaming around in an unknown space, what can you learn about it?
• Two parts of the problem:
• Mapping: As you roam around the world, how do you build a memory of the
shape of the space you have moved through?
• Exploration: Given that you don’t know the shape or size of the environment,
how to make sure you covered all of it?
• Both have many uses:
• Returning back to home/charger after some task.
• Cleaning a new room efficiently; Systematic search for survivors
• Mapping a collapsed mine or building.
• Mapping and Exploration are also “collections of algorithms”
• We will focus on “Occupancy Grid” algorithms

4/28/2021 CSE461 : Introduction to Robotics 21

What is an Occupancy Grid?
•  A way of representing a map as a gridded world where each cell is either “occupied” or “empty” or
“unknown”.

4/28/2021 CSE461 : Introduction to Robotics 22

Examples

4/28/2021 CSE461 : Introduction to Robotics 23

What is a Sensor Model?
• Constructing a Sensor Model
• A sensor measures raw values in an environment
• You have to map that into a Grid Cell Value.
• Robots can have very different sensors and configurations
• Examples:
LIDAR/Depth Camera
Vs. a 360 degree vision/ranging system

4/28/2021 CSE461 : Introduction to Robotics 24

Constructing a Sensor Model
• Example: Depth Sensor Model
R = maximum range, B = maximum angle
Let say the sensor at point p returns distance = “r”
Region 1 (dist < r, grid cell probably empty)
Region 2 (dist = r, grid cell probably obstacle)
Region 3 (dist > r, grid cell unknown/obscured)

• Simplest Sensor Model

Where I stand is Empty (white)

• A Better Model
Set Region 1 cells as Empty (white)
Set Region 2 cells as Occupied (black).
Pick a max range/angle where data is reliable
Rest is still Unknown (gray)

4/28/2021 CSE461 : Introduction to Robotics 25

A Simple OG Mapping Algorithm
1. Initialize a Grid
• Set all locations as “unknown”, pick a start location and orientation
2. Update the Grid
• Mark your current grid position as “empty”
• Using your better sensor model,
• Mark all visible grid locations as “empty” or “occupied”
3. Pick a Next Move
• Look at neighboring grid positions in your map
• Pick a neighboring grid location that is empty (randomly)
• Move to it and update your current position in the Grid
4. Loop forever
• Keep moving and updating the grid (unless you are “done”)

4/28/2021 CSE461 : Introduction to Robotics 26

4/28/2021 CSE461 : Introduction to Robotics 27
Exploration
• Basic Concept in Robotics: Navigating a GRID Graph is
different
• DFS works, but will still make a robot retrace steps
• Better choice: Frontier Based Exploration

4/28/2021 CSE461 : Introduction to Robotics 28

Exploration in Grid Worlds
• Frontier Based Exploration
• A common technique for building maps
• Key Idea:
• Identify the “frontiers” between known and unknown
Frontier cell = a unknown cell with at least one empty cell
• Pick a frontier cell (e.g. the closest)
Plan a path to go explore it.
• Done Condition:
No more frontier nodes left => your map is Complete!
If finite world, then any algorithm that systematically explores frontier
nodes is guaranteed to cover the whole world.

4/28/2021 CSE461 : Introduction to Robotics 29

4/28/2021 CSE461 : Introduction to Robotics 30
4/28/2021 CSE461 : Introduction to Robotics 31
Summary

4/28/2021 CSE461 : Introduction to Robotics 32

Summary

Simultaneous Localization And Mapping

4/28/2021 CSE461 : Introduction to Robotics 33