MCL

Question 1

Continuous vs global localisation

Answer 1

continuous: estimate new position based on estimate of previous and new measurements

global: particles spread randomly through map

Question 2

Steps of MCL

Answer 2

1. motion prediction based on odometry
2. measurement update based on outward looking sensor
3. normalisation
4. resampling

Question 3

How do we do the measurement update for MCL?

Answer 3

apply Bayes rule to each particle

Question 4

How does the likelihood work for sonar update?

Answer 4

• further away measurement from prediction = less likely to occur

Question 5

What is a robust likelihood?

Answer 5

• heavy tail
• constant probability sensors gives garbage distributed across sensor range (add K)
• less aggressive at killing off far measurements

Question 6

How does resampling work?

Answer 6

1. generative cumulative probability distribution
2. generate uniformly distributed random number from 0 to 1
3. pick particle which cumulative probability intersects random number

Question 7

How does a compass sensor work?

Answer 7

• measures bearing relative to north
• to do likelihood do angle measured - bearing