Hello,

I'm trying to follow the tutorial on this website about circle-line collision.

The following is a copy from that tutorial:


[color="#283769"][font="Arial, sans-serif"]Find several other points:[/font]

• The point of intersection between line and the movement vector of the circle. (a)
• The closest point on the line to the endpoint of the movement vector of the circle. (
• The closest point on the movement vector to [color="#006000"][font="monospace"](x1, y1)[/font]. ©
• The closest point on the movement vector to the other endpoint. (d)

[color=#283769][font=Arial, sans-serif]Using the points, it is then possible to eliminate many positions that would not result in a collision. A collision is possible if:[/font]

• a is less than [color="#006000"][font="monospace"]r[/font] away from each endpoint on the movement vector and on the line.
• b is less than r away from the endpoint of the movement vector and on the line segment.
• c is less than r away from [color="#006000"][font="monospace"](x1, y1)[/font] and on the movement vector.
• d is less than r away from [color="#006000"][font="monospace"](x2, y2)[/font] and on the movement vector.

[color=#283769][font=Arial, sans-serif][color="#000000"][font="arial, verdana, tahoma, sans-serif"]Now, my problem is that my tests mostly fail on these last four points even when it is obvious that there is a collision. Take the above image for an example.[/font]
[color="#000000"][font="arial, verdana, tahoma, sans-serif"] [/font]
[color="#000000"][font="arial, verdana, tahoma, sans-serif"]Is this tutorial wrong/incomplete or am I missing something?
[/font][/font]
Thanks in advance!

Based on a single read-through, I can't really confirm whether the approach described in the tutorial is valid. (Maybe it is - it's just not immediately clear to me one way or the other.)

Here however is a method that's guaranteed to produce the correct results (provided all potential numerical issues are handled correctly, of course).

To determine whether a collision will occur, and if so, when it will occur and what the contact point and normal will be, intersect a line segment representing the motion of the circle with a capsule for which the line segment you're testing against is the medial segment, and for which the radius is the circle radius. This approach takes a little work to implement, but it'll handle all cases correctly.

(Again, maybe the method you're using now works, although on first read at least it seems a little questionable. Maybe someone else can comment on that further.)

Wow, thats a great help! Thanks!

I will let you know how it went.

This capsule method works great indeed. Still, at some point when the circle moves very slow the circle goes right through the wall. External forces are applied before the collision detection and do not alter the position of the circle, so that should not be the problem. Any ideas?

This capsule method works great indeed. Still, at some point when the circle moves very slow the circle goes right through the wall. External forces are applied before the collision detection and do not alter the position of the circle, so that should not be the problem. Any ideas?

Can't say for sure without seeing the code, but numerical problems can arise if the magnitude of the displacement is very low. (This problem is unavoidable even from a purely mathematical standpoint, since it makes little sense to compute the time of intersection for a non-moving object, unless that time is zero of course.)

I wouldn't expect the test to fail though unless the magnitude of the displacement vector was very low or (potentially) the circle was moving parallel or nearly parallel to the capsule axis. If you're finding that the test is failing in some cases, I'd recommend doing some debugging to try and determine where exactly the failure is occurring. (It may be due to an error in your implementation, or simply due to not handling the various potential numerical issues correctly.)