Obstacles

An obstacle is represented by a MOG. Each member is a possible location for the obstacle.

First Approach to representing obstacle

If there is a cylindrical obstacle of width w, and the obstacle can be placed at location (x,y) where y varies between y1 and y2, then the resulting obstacle is most simply represented as a single normal distribution with mean=(x,y) and with XX variance of w^2 and YY variance of ((y2-y1)*w)^2.

Avoid obstacles checks for overlap between the MOG of the robot location and that of the obstacle. If any member pair ((i,j) with i in robot pos MOG and j in obstacle MOG) has means within 1m of each other, then overlap is calculated.The 1M comes from the CNL 1m parameter used for obstacle avoidance.

overlap is calculated using the Bhattacharya measure, and the distribution resulting from this is the sensed positioned of the obstacle.

The calculation of obstacle avoidance is as done in the Avoid_Obstacles process as
BaseSpeed*(1 - max(ObstacleDistance/1m,1))

Issues: 1. This conflates the physical geometry with the position uncertainty. 2. The 1m overlap is only wth respect to means -- thus an 'elongated' variance is ignored 3. The 1m should not be built in; its part of the obstacle avoidance method

Modification 1

The 1m parameter is dispensed with and instead a significance test is used. The distance between means in a pair is compared to the following value
sqrt( pow( sqrt(XX1)+sqrt(XX2) , 2) +powe( sqrt(YY1)+sqrt(YY2) , 2) )

This distance is a 1SD significance test; if the distances are closer than this, the overlap is evaluated. This fixes two things a. an alongated variance will result in the obstacle being 'seen' earlier b. this eliminated the 1m strategy parameter from the world modelling part of the problem

-- (c) Fordham University Robotics and Computer Vision