Probabilistic Methods in Mobile Robotics
Stereo cameras Infra-red Sonar Laser range-finder Sonar Tactiles
Bayes Formula
A Simple Example: Estimating the state of a door u Suppose a robot obtaines measurement s u What is p(Door=open|SensorMeasurement=s) ? u Short form: p(open|s)
Causal vs. Diagnostic Reasoning u We’re interested in p(open|s) (called diagnostic reasoning) u Often causal knowledge like p(s|open) is easier to obtain. u From causal to diagnostic: Apply Bayes rule:
Normalization
Example u p(s|open) = 0.6p(s| open) = 0.3 u p(open) = p( open) = 0.5 s raises the probability, that the door is open.
Integrating a second Measurement... u New measurement s 2 u p(s 2 |open) = 0.5p(s 2 | open) = 0.6 s 2 lowers the probability, that the door is open.
Where am I? + Mobile Robot Localization
Principle of Robot Localization
l L t : position of the robot at time t l Given: l Map and sensor model: l Motion model: l Initial state of the robot: l Data Sensor information (sonar, laser range-finder, camera) o i Odometry information a i Markov Localization as State Estimation (1)
Motion Model
Model for Proximity Sensors l The sensor is reflected either by a known or by an unknown obstacle : Laser sensor Sonar sensor
Motion: Perception: … is optimal under the Markov assumption Kalman filters, Hidden Markov Models, DBN Markov Localization as State Estimation (2)
Grid-based Markov Localization Three-dimensional grid over the sate space of the robot:
Localization Example (1)
Localization Example
Sample-based Density Representation D. Fox, Univ. of Washington
Global Localization (sonar)
Example Run Sonar
Example Run Laser
Localization for AIBO robots D. Fox, Univ. of Washington
Localization for AIBO robots D. Fox, Univ. of Washington
Mobile Robot Mapping
Mapping the Allen Center: Raw Data
Mapping the Allen Center
Multi-robot Mapping Robot ARobot BRobot C