Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Scenario issues #2004

Closed
Adaickalavan opened this issue Apr 29, 2023 · 8 comments · Fixed by #2010
Closed

Scenario issues #2004

Adaickalavan opened this issue Apr 29, 2023 · 8 comments · Fixed by #2010
Labels
bug Something isn't working

Comments

@Adaickalavan
Copy link
Member

Adaickalavan commented Apr 29, 2023

High Level Description

This is on the merge_exit map. Expected waypoints to behave like that shown here. But got the below issues in chronological order:

  • In on-ramp, agent sees three waypoint paths. [ok]
  • In on ramp, three waypoint paths extends across the junction. [ok]
  • In junction, agent sees single waypoint path which jumps around. [probably scenario issue]
  • After merge in highway, waypoint paths end abruptly. [probably scenario issue]
  • In highway, three short and thick lines appear. [probably scenario issue]
  • In highway, at the very end of the gif, obs.waypoint_paths is empty, causing the episode to crash. [probabaly scenario issue]

Peek 2023-04-28 19-40

Version

branch: master
version: v1.0.11

Steps to reproduce the bug

Replay the `merge_exit` scenario and visualize the waypoints.

System info

No response

Error logs and screenshots

No response

Impact (If known)

The merge_exit scenario is unusable.

@Adaickalavan Adaickalavan added the bug Something isn't working label Apr 29, 2023
@qianyi-sun
Copy link
Contributor

@Adaickalavan The issues are mostly due to overlapped edges around the junctions, if the edges are overlapped, the waypoints are overlapped as well, which leads to:

  • confusion on picking the right waypoints path (jump around issue)
  • dense, crowded waypoints in a small edge (three thick lines)

To fix this,

  • modified the geometry endpoints for most of the edges to minimize the overlapping around the junction so that the waypoints are generally clearer and smoother.
  • extend edges/junctions a bit for same reason above, and also can give more room/time to access the upcoming waypoints in the observations

Please take a look at the following gif and let me know if this is acceptable, I have already pushed this map onto the repo, thanks!

merge_fix

@Adaickalavan Adaickalavan changed the title Waypoints and scenario issues Scenario issues Apr 29, 2023
@Adaickalavan
Copy link
Member Author

More scenario issues on straight map.

straight
Peek 2023-04-28 22-15

  • Use entry tactic method to ensure lead vehicle appears before ego, rather than map design methods.
  • Check and rectify all scenarios for similar issues.

@Adaickalavan
Copy link
Member Author

The modified merge_exit map seems to use junctions with lane-change connections to specify the locations where the lead vehicle could change lane, albeit the decision to change lane is random. Since the above gif uses an EndlessMission agent, the same set of waypoints will be seen by the ego agent. Essentially, the modified maps reveal a degree of privileged info to the ego on where potential lane changes by the lead vehicle could happen, in contrary to the idea of the task where ego should not have any prior knowledge of the leader's trajectory.

Screenshot from 2023-04-29 10-33-11

Screenshot from 2023-04-29 10-51-50

@qianyi-sun
Copy link
Contributor

More scenario issues on straight map.

straight straight Peek 2023-04-28 22-15 Peek 2023-04-28 22-15

  • Use entry tactic method to ensure lead vehicle appears before ego, rather than map design methods.
  • Check and rectify all scenarios for similar issues.

We haven implemented such entry tactic yet, i believe it's in progress on PR#1965.

@Adaickalavan
Copy link
Member Author

On the modified merge_exit map, we see the following:

  • The lead vehicle waits too long at the merge such that the episode terminates due to reaching max_steps_per_episode.
  • The background traffic seems to always change lane to left-most lane prior to the merge location. This severely reduces any traffic interaction this scenario presents to the ego.
  • Background traffic vehicles and lead vehicle seems to akwardly gain speed between the newly introduced consecutive lane-change junctions in the main road.

Peek 2023-04-29 11-43

@qianyi-sun
Copy link
Contributor

qianyi-sun commented Apr 29, 2023

On the modified merge_exit map, we see the following:

  • The lead vehicle waits too long at the merge such that the episode terminates due to reaching max_steps_per_episode.
  • The background traffic seems to always change lane to left-most lane prior to the merge location. This severely reduces any traffic interaction this scenario presents to the ego.
  • Background traffic vehicles and lead vehicle seems to akwardly gain speed between the newly introduced consecutive lane-change junctions in the main road.

Peek 2023-04-29 11-43 Peek 2023-04-29 11-43

  • re 1): There can be three possible solution: 1) reduce the traffic density 2) give the on-ramp higher road priority, which would lead to all traffic vehicle stops at the intersection when the leader is merging in. 3) put a traffic light on that junction.

Please let me know which one is prefered, imp, it depends on how we see this scenario as, if we see this as a hiway scenario, then reducing density might be appropriate. if it's just a urban road with some merging junctions, we might want to add traffic light or modify the road right priority instead.

  • re 2): I dont see this happens on my end, it's pretty random as shown below
  • re 3): I reported this to @Gamenot already, we checked every speed config/info in those files, the cause is unknown and might be a sumo issue.

merge_traffic

@qianyi-sun
Copy link
Contributor

I have pushed several map changes and tested them, please take a look and let me know if any more change is required, thanks @Adaickalavan !

@qianyi-sun
Copy link
Contributor

I reduced the traffic density a bit, the current traffic looks reasonable to me and doesnt block the leader to merge in

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants