Skip to content

Latest commit

 

History

History
35 lines (29 loc) · 2.24 KB

README.md

File metadata and controls

35 lines (29 loc) · 2.24 KB
Raw

RoboFieldLocalisation

Using the video as found in https://youtu.be/DYQFtrk6B18.

Goal

  • Find an approximate location for the robot in the field (Note: Because of field symmetry there are likely to be more possible locations)

How?

(step 1)

Find robot's position in respect of the visible lines from the robot's vision (circle in the video):

  • Find the lines
  • Retrieve a random set of points on the visible lines (pts)
  • Compensate lens distortion for each pt in pts
  • Calculate values comparable to the real world distance and angle for each pt in pts

(step 2).

Localization in total field:

  • Use n particles (prs), uniformly distributed over the field
  • Move each pr in prs in an random direction (in each iteration)
  • Take, for each particle a random set of points (pts) on lines within x radius from the particle
  • Calculate values comparable to the real world distance and angle for each pt in pts
  • Calculate (for this iteration) the difference between the set of (distance, angle) to the set (distance, angle) retrieved from the robot's vision (in step 1)
  • The latest calculated value is the value we're optimizing on. If small enough, slow the particle (and TODO: change direction towards the local optimum). If large, speed up.
  • Any particles going out of bounds will be removed and placed randomly near an existing particle or anywhere on the field.
  • Particles in the field will also randomly be replaced every once in a while. This prevents only finding local optima

Issues:

  • Parameters: How many particles, how often do we replace? What is the difference between distances in the video and on the field.
  • The optimization parameter in step 2. How do we calculate this? Using distances and angles we should get near. But do we have enough random point selections to get close?
  • The optimization parameter heavily depends on the first issue. The difference between distances in the field and seen from the robot
  • The resulting lines as seen from the robot are distortion free. Not yet perspective free. This should make a large difference.

Sample image

Sample image of what it looks like. At a given early moment in the process Screenshot