Sensors Fusion for Mobile Robotics localization
Until now we’ve presented the main principles and features of incremental and absolute (environment referred localization systems) could you summarize the main features and differences??? Main problems of both categories??? Need more/better pictures
Robot Sensors - localization Infrared Ranging Do you recognize the difference between the two categories? Magnetometer GPS IR Modulator Receiver Accelerometer Camera Linear Encoder Sonar Ranging Gyroscope Rotary Encoder Laser triangulation Laser Rangefinder Need more/better pictures Compass Incremental vs Absolute
Robot Sensors - localization Good features Bad features Need more/better pictures Incremental vs Absolute
Example: localization with encoders Example: the vehicle shall be driven on a corridor localized only by encoders mounted on the wheels. Problem: Left wheel smaller radius (wrt to the nominal value). Ideal = the path that the vehicle assumes to lie on Drift
Importance of Uncertainty Estimation By estimating the uncertainty it is possible to detect and avoid accident but also to combine information Importance of Uncertainty Estimation
Uncertainty Estimation – Sensor Fusion Event: vehicle localization with another sensor referred to the environment (for example a laser triangulation, a camera, etc) Uncertainty Estimation – Sensor Fusion
Without using uncertainty: simple average Uncertainty Estimation – Sensor Fusion
Uncerainty Estimation – Sensor Fusion Using uncertainty: Sensor Fusion Uncerainty Estimation – Sensor Fusion
Questions?
Incremental vs Global Localization Vehicle localization main classification : INCREMENTAL LOCALIZATION The current vehicle pose at time t is evaluated wrt the information achieved in the previous localization at time t-1. GLOBAL LOCALIZATION The current vehicle pose at time t is evaluated wrt the information referred to a global reference system. Incremental vs Global
Incremental an Global Localization Σ0: global reference system Vehicle is globally localized with a direct estimation of H0,k. Vehicle is incrementally localized using the concatenation of the estimations Hi,j. Incremental vs Global
Incremental an Global Localization Incremental vs Global
INCREMENTAL LOCALIZATION Sensor (most used one) classification: INCREMENTAL LOCALIZATION GLOBAL LOCALIZATION Encoders on wheels Triangulation Systems Gyroscope + magnetometers Ultrasound beacon Laser Scanner – comparison with previous acquisition with a map Camera looking on the floor Camera looking on the ceiling Incremental vs Global
Incremental an Global Localization Feature INCREMENTAL GLOBAL Drift in pose estimation HIGH NO Measurement update rate LOW Repeatability Needs of environment information YES Incremental vs Global
Odometric - Global Navigation Fusion First issue: time alignment due to the different update rate Incremental-Global Localization Sensor Fusion
Odometric - Global Navigation Fusion First issue: time alignment due to the different update rate Incremental-Global Localization Sensor Fusion
Odometric - Global Navigation Fusion Second issue: Sensor Fusion and how to continue! Example Matlab Incremental-Global Localization Sensor Fusion
Incremental and Global Localization HIGH, SMOOTH TRAJECTORY Feature INCREMENTAL GLOBAL SENSOR FUSION Drift in pose estimation HIGH NO Measurement update rate LOW Repeatability HIGH, SMOOTH TRAJECTORY Needs of environment information YES Incremental-Global Localization Sensor Fusion
Example: Use of encoders + gyro + laser triangulation … my first industrial AGV
Sensor Fusion Laser triangulation Encoders Gyro No drift 1° STEP (a) 1° STEP (b) Laser triangulation Encoders Gyro 2° STEP No drift Low repeatability (especially in motion or with low number of reflectors) High frequency of update Drift 1° & 2° STEP: High frequency of update & No drift Sensor Fusion
1° STEP (a) Fusion between incremental systems xR calibrated as a Function of the manoeuvre * Fusion of the increments Kinematics equations … already seen this example
1° STEP (b) Real time covariance estimation X is the POSE (position and attitude) * White noise This part takes into account correlation as a function of time wk vector of the uncertainty parameters
2° STEP (a) Estimation of covariance of laser triangulation as a function of the manoeuvre 1. State of the encoders 2. Laser quality factor * 2° STEP (b) Fusion between environment referred and incremental estimations
C.I. 2 sigma C.I. 30 sigma
Delay
List of symbols:
X is the pose (position and attitude)