Source: SLAM for Dummies

Random Sampling Consensus

• to extract lines from a laser scan
• lines then used as landmarks
• indoors: straight lines from walls
• line landmarks are found by
  • randomly taking a sample of laser readings (e.g. sample readings from 12deg to 22deg from within a range of 0 to 180deg)
  • least squares approximation for line of best fit
  • RANSAC then checks how many laser readings lie close to the best fit line
    • initially, all readings are assumed to be unassociated to any lines
    • if the num. of readings > threshold, then a line has been seen
    • this threshold is the consensus

The EKF  implementation assumes that landmarks come in as single points with range and bearing [r, b] from the robots position

Translate a line to a fixed point:

• Take a fixed point in the world coordinates
• Calculate point on the line closest to the fixed point
• Using the robot’s position and the position of the FP on the line, a range and bearing can be calculated

Another possibility: EKF handles lines instead of just points