Skip to content

Decide follow/pass per object prediction in the node - #36

Draft
jpbusch wants to merge 3 commits into
mainfrom
follow-pass
Draft

jpbusch wants to merge 3 commits into
mainfrom
follow-pass

Conversation

@jpbusch

@jpbusch jpbusch commented Nov 16, 2026

Copy link
Copy Markdown
Member

Each object prediction is either passed or followed. The decision is made in the node (new class FollowPass) instead of implicitly in the OCP:

  • passed predictions bound the lateral offset from the reference path (soft, with d_min_object_lat or a per-class margin)
  • per stage, the closest followed prediction in the suggested lane gives an arc-length follow distance (soft d_min_object_long and thw, hard d_min_object_collision); the soft part is relaxed to what full braking can still reach
  • the initial guess starts from full braking if that is needed to keep d_min_object_long; a failed constrained solve is retried from the other initial guess within the remaining timeout

The hard SAT separation stays for all other object slots. The thw slack weight is coupled to the lateral cost weight (w_lat / c_lat^2).

jpbusch and others added 3 commits November 16, 2026 11:06
Each object prediction is either passed or followed. The decision is made
in the node (new class FollowPass) instead of implicitly in the OCP:

- passed predictions bound the lateral offset from the reference path
  (soft, with d_min_object_lat or a per-class margin)
- per stage, the closest followed prediction in the suggested lane gives an
  arc-length follow distance (soft d_min_object_long and thw, hard
  d_min_object_collision); the soft part is relaxed to what full braking
  can still reach
- the initial guess starts from full braking if that is needed to keep
  d_min_object_long; a failed constrained solve is retried from the other
  initial guess within the remaining timeout

The hard SAT separation stays for all other object slots. The thw slack
weight is coupled to the lateral cost weight (w_lat / c_lat^2).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant