Introduction to Mobile Robotics Probabilistic Sensor Models

Transcription

1 Introduction to Mobile Robotics Probabilistic Sensor Models Wolfram Burgard 1

2 Sensors for Mobile Robots Contact sensors: Bumpers Proprioceptive sensors Accelerometers (spring-mounted masses) Gyroscopes (spinning mass, laser light) Compasses, inclinometers (earth magnetic field, gravity) Proximity sensors Sonar (time of flight) Radar (phase and frequency) Laser range-finders (triangulation, tof, phase) Infrared (intensity) Visual sensors: Cameras Satellite-based sensors: GPS

3 Proximity Sensors The central task is to determine P( x), i.e., the probability of a measurement given that the robot is at position x. Question: Where do the probabilities come from? Approach: Let s try to explain a measurement. 3

4 Beam-based Sensor Model Scan consists of K measurements. =,,..., { 1 K } Individual measurements are independent given the robot position. P( x, m) = K k = 1 P( k x, m) 4

5 Beam-based Sensor Model P( x, m) = K k = 1 P( k x, m) 5

6 Typical Measurement Errors of an Range Measurements 1. Beams reflected by obstacles. Beams reflected by persons / caused by crosstalk 3. Random measurements 4. Maximum range measurements 6

7 Proximity Measurement Measurement can be caused by a known obstacle. cross-talk. an unexpected obstacle (people, furniture, ). missing all obstacles (total reflection, glass, ). Noise is due to uncertainty in measuring distance to known obstacle. in position of known obstacles. in position of additional obstacles. whether obstacle is missed. 7

8 Beam-based Proximity Model Measurement noise Unexpected obstacles 0 exp max 0 exp max P hit ( x, m) = η 1 πb e 1 ( b exp ) P unexp ( x, m) = λ η λ e < exp 0 otherwise 8

9 Beam-based Proximity Model Random measurement Max range 0 exp max 0 exp max P rand ( x, m) = η 1 max P max ( x, m) = η 1 small 9

10 10 Resulting Mixture Density = ), ( ), ( ), ( ), ( ), ( rand max unexp hit rand max unexp hit m x P m x P m x P m x P m x P T α α α α How can we determine the model parameters?

11 Raw Sensor Data Measured distances for expected distance of 300 cm. Sonar Laser 11

12 Approximation Maximie log likelihood of the data Search space of n-1 parameters. Hill climbing Gradient descent Genetic algorithms P( exp Deterministically compute the n-th parameter to satisfy normaliation constraint. ) 1

13 Approximation Results Laser Sonar 300cm 400cm

14 Example P( x,m) 14

15 Discrete Model of Proximity Sensors Instead of densities, consider discrete steps along the sensor beam. Laser sensor Sonar sensor 16

16 Approximation Results Laser Sonar 17

17 Influence of Angle to Obstacle "sonar-0"

18 Influence of Angle to Obstacle "sonar-1"

19 Influence of Angle to Obstacle "sonar-"

20 Influence of Angle to Obstacle "sonar-3"

21 Summary Beam-based Model Assumes independence between beams. Justification? Overconfident! Models physical causes for measurements. Mixture of densities for these causes. Assumes independence between causes. Problem? Implementation Learn parameters based on real data. Different models should be learned for different angles at which the sensor beam hits the obstacle. Determine expected distances by ray-tracing. Expected distances can be pre-processed.

22 Scan-based Model Beam-based model is not smooth for small obstacles and at edges. not very efficient. Idea: Instead of following along the beam, just check the end point. 3

23 Scan-based Model Probability is a mixture of a Gaussian distribution with mean at distance to closest obstacle, a uniform distribution for random measurements, and a small uniform distribution for max range measurements. Again, independence between different components is assumed. 4

24 Example Likelihood field Map m P( x,m) 5

25 San Jose Tech Museum Occupancy grid map Likelihood field 6

26 Scan Matching Extract likelihood field from scan and use it to match different scan. 7

27 Properties of Scan-based Model Highly efficient, uses D tables only. Distance grid is smooth w.r.t. to small changes in robot position. Allows gradient descent, scan matching. Ignores physical properties of beams. 9

28 Additional Models of Proximity Sensors Map matching (sonar, laser): generate small, local maps from sensor data and match local maps against global model. Scan matching (laser): map is represented by scan endpoints, match scan into this map. Features (sonar, laser, vision): Extract features such as doors, hallways from sensor data. 30

29 Landmarks Active beacons (e.g., radio, GPS) Passive (e.g., visual, retro-reflective) Standard approach is triangulation Sensor provides distance, or bearing, or distance and bearing. 31

30 Distance and Bearing 3

31 Probabilistic Model 1. Algorithm landmark_detection_model(,x,m): = i, d, α, x = x, y,θ. 3. dˆ = ( m x ( i) x) + ( m y ( i) y) ˆ α = atan( m ( i) y, m ( i) x) θ y x 4. p det prob( dˆ d, ε ) prob( ˆ α α, ε ) = d α 5. Return p det 33

32 Distributions 34

33 Distances Only x = ( a + d 1 d ) / a No Uncertainty y = ± ( d 1 x ) X a P 1 P d 1 d x P 1 =(0,0) P =(a,0) 35

34 Bearings Only No Uncertainty P α P D P 3 β D 3 α 3 D 1 D 1 P 1 1 α P 1 1 D Law of cosine 1 = cosα D D D 1 3 = = = cos( α) cos( β ) cos( α + β ) 36

35 Bearings Only With Uncertainty P P 3 P P 1 P 1 Most approaches attempt to find estimation mean. 37

36 Summary of Sensor Models Explicitly modeling uncertainty in sensing is key to robustness. In many cases, good models can be found by the following approach: 1. Determine parametric model of noise free measurement.. Analye sources of noise. 3. Add adequate noise to parameters (eventually mix in densities for noise). 4. Learn (and verify) parameters by fitting model to data. 5. Likelihood of measurement is given by probabilistically comparing the actual with the expected measurement. This holds for motion models as well. It is extremely important to be aware of the underlying assumptions! 38