7b - SLAM Flashcards
What is SLAM an example of?
Chicken or egg problem
If we can’t know where we are, we cannot make a good model. But if we don’t have a good model we can’t know where we are
What does SLAM stand for?
Simultaneous Localisation and Mapping
What is the SLAM problem?
How can a body navigate in a previously unknown environment, while constantly building and update a map of its workspace using onboard sensors and onboard computation
When is SLAM necessary?
When a robot must be truly autonomous (no human input)
When there is no prior knowledge about the environment
When we cannot rely exclusively on pre-placed beacons or external positioning systems
When the robot needs to know where it is
What are the steps of SLAM with a Gaussian Filter?
On every frame, predict how the robot has moved, measure, update the internal representations
What is the process of updating the internal representations?
The robot observes a feature which is mapped with an uncertainty related to the measurement model
e.g the camera model describing how world points map into pixels in the image
What is the process of predicting how the robot has moved?
As the robot moves, its pose uncertainty increases, obeying the robot’s motion model
e.g the driver’s commands, uncertainty is added due to wheel slippage and other imprecisions
What is the process of measuring?
Robot observes two new features, then their position uncertainty results from the combination of the measurement error with the robot pose uncertainty
Map becomes correlated with the robot pose estimate
What is the different between EKF SLAM and Particle Filter SLAM?
Standard EKF SLAM represents the probability distribution in parametric form with a Gaussian
Particle filter SLAM represents the probability distribution as a set of particles drawn randomly from the parametric distribution
What is the update process for particle filter SLAM?
Position uncertainty is encoded for each particle individually
Compare the particles predicted measurements with actual measurements
Re-weight s.t particles with good predictions getting higher weight
Renormalise particle weights
Resample
How does SLAM with a particle filter represent beliefs?
With a series of samples
Each particle denotes a hypothesis of the state with an associated weight
Predict/measure/update
What are the pros of SLAM with a particle filter?
Noise densities from any distribution
Works for multimodal distributions
Easy to implement
What are the cons of SLAM with a particle filter?
Does not scale to high dimensional problems
Requires many particles to have good convergence
