State Estimation
Estimate the state of x of a system given obesrvatioins z and controls u
Recursive Bayes Filter
Define belief
With Bayes rule
With Markov assumption:
With law of total probability
With again Markov assumption
With the assumption the current state $ x_{t} $ is independent of the future controls $ u_{t+1} $
A recursive term!
Prediction and correction step
Prediction step
where $ p(x_t \vert x_{t-1}, u_t) $ is so-called the motion model
Correction step
where $ p(z_t \vert x_t) $ is the observation model.
Different Realizations
Framework for recursive state estimation
Different relaization and properties
- Linear vs. non-linear models for motion/observation models
- Gaussian distribution only?
- Parametric filters
Kalman filter & friends
- Gaussians
- Linear or linearized models (note: motion is often non-linear because of rotations)
Particle fileter
- Non-parametric
- Arbitrary
Robot Motion models
- Robot motion is inherently uncertain, How to model uncertainty?
Probabilistic Motion Models
Specify a posterior probabiilty that carries the robot from x to x’:
Odometry-based
Standard odometry model
#TODO the nice diagram
Probability Dist
- Noise in $ (x, y, \theta) $, e.g. $ u \sim N(0, \Sigma) $
#TODO the nice diagram and examples
Velocity-based
$ (x, y, \theta) $ to $ (x’, y’, \theta’) $ with translation velocity v and rotation velocity w.
- Problem: No arc could describe $ \theta’ \neq \theta $
- Fix: Extra term for the final rotation
Sensor Model
Model for Laser Scanners
Measure Time-of-Flight
Scan z consists of k measurements
Assumption: Individual measurements are independent given the robot position
Beam-Endpoint Model
Physically flaw bu_t efficient
Ray-cast Model
Mixture of four models: exponential decay in distance, gaussian at obstacle, delta at the max-distance and a uniform distribution.
Model for perceiving landmarks with Range-Bearing Sensors
Range-Bearing
Robot’s pose
Observation of feature j at location $ (m_{j,x}, m_{j,y}) $
Reading materials
Bayes filter
- PR Chapter 2
- Introduction to Mobile Robotics Chapter 5
Motion and sensor models
- PR 5 & 6
- Course 6 & 7