search

elvout / cs393r

CS 393R Graduate Autonomous Robots, Fall 2021 | Autobots
0 stars 0 forks source link

Arc selection heuristics #9

Closed aiddun closed 3 years ago

aiddun commented 3 years ago

Given sensor readings and a path trajectory, calculate

elvout commented 3 years ago

I'm thinking that we don't really need to create a function mapping free path length and clearance to a score. My intuition is that all paths we consider just need to have a clearance greater than some threshold value.

aiddun commented 3 years ago

In that case what if we just made the car really wide in our calculations?

elvout commented 3 years ago

I believe our calculations already do something like that – we have a safety margin around the car. However, by design, the code does not distinguish between "real" collisions that actually hit the car and "safety" collisions that occur within the safety margin. I think we should still calculate clearance for each path and then filter out paths with small clearance when we find the best path to take.

elvout commented 3 years ago

We also have an edge case where there's just a long wall separating the car from its target


                 |
                 |
                 |
->               |             x
                 |
                 |
                 |

Our algo decides to just go straight since most of the paths terminate at the wall.

Maybe this should be moved it its own issue.

aiddun commented 3 years ago

  1. If we're filtering out anyway, why not just make the car super wide then?

  2. Do we want forward clearance too then? It's a bit of a weird concept but if a car can't turn once it reaches the max point it can go due to an obstacle

On Fri, Sep 17, 2021 at 11:31 AM Elvin Yang @.***> wrote:

We also have an edge case where there's just a long wall separating the car from its target

             |
             |
             |
-> x

Our algo decides to just go straight since most of the paths terminate at the wall.

Maybe this should be moved it its own issue.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/elvout/cs393r-a1/issues/9#issuecomment-921929087, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACOQ5KAHRPLGEFWCTR3CV7TUCNUOZANCNFSM5DYAL2YA .

elvout commented 3 years ago
  1. If we're filtering out anyway, why not just make the car super wide then?

That's a good point. Actually, after refactoring the code and removing an assertion, the code no longer crashes when there is a collision on the safety margin.

  1. ...

I was thinking we factor in free path length into a scoring function. However, this might cause the car to take a longer path even when it shouldn't. We an look into some sort of forward clearance.