6 - Localisation & Mapping Flashcards
Name 3 methods of localisation
Based on external sensors, beacons and landmarks
Odometry (proprioceptive sensors)
Map-based (no external sensors and landmark, only onboard sensors)
Discuss GPS positioning
GPS provides metre-level outdoor positioning using information from multiple satellites
What are problems with GPS?
Atmospheric conditions limit precision unless corrected using a reference
Can be power hungry
Limited accuracy
Poor in urban environments
Need infrastructure
Discuss beacon based localisation
Generally limited accuracy
External references can be used for triangulation
City line of sight may be blocked
Signals may be jammed
What is a Home vector?
Estimated location = vector relative to home
What is Odometry?
An approximation of the location of a robot can be obtained by repeatedly computing the distance moved and the change direction from the velocity of the wheels in a short period of time
What is used to try and reduce the accumulated errors in Odometry?
Heading sensors, such as a gyroscope
What is Visual Odometry?
Odometry is not limited to wheel encoders, in visual odometry consecutive camera images are used to estimate velocity
What is the solution to the problem of self-localisation?
Use a map. Combine odometry with sightings of known landmarks/environmental features
Name 5 sources of Odometry errors
Misalignment of wheels
Unequal wheel diameter
Variation in the contact point
Unequal floor contact of the wheel
Limited resolution during integration
What are the 3 types of Odometry error?
Range, turn and drift
What is a Range rrror?
Sum of the wheel motion leads to an error in the integrated distance of the robot’s movement
What is a Turn error?
Difference of the wheel motions leads to an error in the robot’s final orientation
What is a Drift error?
Difference in the error of the wheels leads to an error in the robot’s angular orientation
What does SLAM provide?
The best of both worlds by using Odometry for rapid position updates which are realigned periodically using the map
What is a map?
A collection of elements or features at some scale of interest and a representation of the spatial and semantic relationships among them
What can maps be used for?
Localisation, planning, mobile manipulation and human-robot interaction
Why is mapping highly labour intensive?
Exploration (global coverage), Measurement, Validity, Currency
Name 4 types of maps
Metric, topological, semantic, hybrid
What is a Metric map?
Records the location of objects in an absolute coordinate system
What is a Topological map?
Records the connections (edges) between a set of places (nodes)
What is a Semantic map?
Records semantic information (metadata), includes segmentation, place/object naming, function
What is a Hybrid map?
Combines two or more map types
What are 2 types of Metric map?
Continuous/vector format and discrete/raster format
What is a continuous/vector metric map?
Points, linear or curved segments, surface patchess
What is a Discrete/raster metric map?
Occupancy grids
What is an Occupancy Grid an example of?
Metric Map
What is an Occupancy Grid?
Where each cell is associated with a probability that the cell is occupied. It’s updated using the current robot pose and depth measurements from range-finding sensors
What are the 4 regions in an Occupancy Grid?
Probably occupied
Probably empty
In the shadow of the detected object so status unknown
Outside the beam so status unknown
What is Probabilistic robotics?
Explicit representation of uncertainty using the calculus of probability theory
What can the edges in a Topological map also represent?
Actions needed to get from one node to the next, or direction and distance
Discuss path planning in Topological maps
Can determine a route by standard graph search methods such as A* or Dijkstra, if the distances or costs are added to the edges
What are 5 advantages of Metric Maps over Topological Maps?
Detailed, quantitative, sub-symbolic representation
Good for representing and avoiding known, static obstacles
High computational cost of storage and processing
Requires very accurate position tracking (reliance on accurate odometry and range findder sensors)
How to determine an appropriate resolution?
What are 5 advantages of Topological maps over Metric Maps?
Abstract, qualitative, symbolic representations
May be more persistent/robust to environment dynamics
Low computational cost (efficient path planning, scale better to large environments)
Require accurate place recognition (problem of perceptual aliasing, what if 2 or more places look alike)
How to determine what makes a place?