Repository navigation
feat(map_based_prediction): suppress undetected lane-change paths when a lane-follow path exists (cherry-pick autowarefoundation#13525) - #3637
Open
YoshiRi wants to merge 1 commit into
Conversation
…n a lane-follow path exists (autowarefoundation#13525) * feat(map_based_prediction): suppress undetected lane-change paths when a lane-follow path exists Vehicles judged as LANE_FOLLOW no longer get the fixed-weight (0.3) left/right lane-change branches; only the lane-follow path is emitted. Lane-change paths are still generated when lateral motion has been observed (LANE_CHANGE_* judgement), when the current lane ends within speed x lane_end_rescue_time_horizon (merge / lane drop), or when no lane-follow path can be built. Suppressed candidates are restored if no lane-follow path survives the later conversion and lateral-acceleration checks, so an object never loses all map-based paths. Downstream, collision_check_filter evaluates every predicted path, and the undetected lane-change branches of neighboring vehicles were the main source of false DANGER/FATAL assessments in urban replays (26 -> 12 segments, no new ones, on 8 recorded Odaiba scenarios). New parameters under lane_change_path_suppression: enable (true), keep_when_lane_ends (true), lane_end_rescue_time_horizon (5.0 s). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Signed-off-by: yoshiri <yoshiyoshidetteiu@gmail.com> * style(pre-commit): autofix * refactor(map_based_prediction): split lane-change path suppression helpers to reduce method complexity No behavior change. Addresses the CodeScene findings on autowarefoundation#13525: - getPredictedReferencePath (212 lines) is split into makeReferencePathSearchContext, calcRoutingCostLimit, getPathsForLanelet, getLeftOrRightLanelet, collectLaneletReferencePaths, updateFuturePossibleLanelets and logLaneChangeSuppressionDecisions; the per-lanelet candidate generation no longer lives in lambdas capturing a dozen locals. - predictOnLane's conversion loop is moved to generateOnLanePaths with an OnLanePathCandidates accumulator. - The lane_change_path_suppression parameter declarations move out of the node constructor into declare_lane_change_path_suppression_params. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Signed-off-by: yoshiri <yoshiyoshidetteiu@gmail.com> * fix(map_based_prediction): fix build and cpplint findings after the refactor - getPathsForLanelet cannot be const: isIsolatedLanelet takes a non-const RoutingGraphPtr reference. - Format the lanelet id with std::to_string instead of a C long cast (cpplint runtime/int). - Include <utility> in test_path_processor_suppression.cpp for std::make_pair. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Signed-off-by: yoshiri <yoshiyoshidetteiu@gmail.com> * fix(map_based_prediction): suppress and restore lane-change paths per origin lanelet - Suppress an undetected LC candidate only when its own current lanelet has a lane-follow candidate; a current lanelet without a center keeps all of its candidates instead of relying on a global "any center exists" flag. - Restore held LC candidates only for origin lanelets whose lane-follow prediction (two or more reference points) was not accepted; origins whose center survived keep their suppression. - Keep the original handling of one-point lane-follow references: the straight path of the path generator is emitted as before, so enable=false (or nothing suppressed) reproduces the previous behavior exactly. - Record the origin lanelet in PredictedRefPath and the per-origin center existence in LaneChangePathSuppressionInfo; update DEBUG decisions, tests and README accordingly. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Signed-off-by: yoshiri <yoshiyoshidetteiu@gmail.com> * docs(map_based_prediction): expand CCF/LC abbreviations and merge duplicated brief comment - README: spell out CCF (Collision Check Filter) and LC (lane change) where they first appear in the suppression section. - map_based_prediction_node.cpp: move the deceleration-parameter brief back onto declare_object_deceleration_params; it had been left above the new lane-change suppression helper. Signed-off-by: yoshiri <yoshiyoshidetteiu@gmail.com> Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> --------- Signed-off-by: yoshiri <yoshiyoshidetteiu@gmail.com> Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com> Co-authored-by: pre-commit-ci-lite[bot] <117423508+pre-commit-ci-lite[bot]@users.noreply.github.com> (cherry picked from commit 2cee537)
|
Tick the box to add this pull request to the merge queue (same as
|
|
Author
|
@mkquda Could you review this PR when you have time? It is a cherry-pick of autowarefoundation#13525 (already merged upstream and in |
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Cherry-pick of autowarefoundation#13525 (merged as 2cee537) onto
feat/v0.64/e2e.map_based_predictionemitted left/right lane-change branches with a fixed weight of 0.3 for every vehicle, even without any observed lateral motion.collision_check_filterevaluates every predicted path, so these branches of neighboring and rear vehicles made the ego vehicle yield to cars that are driving straight (false DANGER/FATAL; 15 of the 26 HIGH/FATAL segments in the Odaiba replay).With this change a vehicle judged as
LANE_FOLLOWemits only its lane-follow path. Lane-change paths are still generated when lateral motion has been observed (LANE_CHANGE_*judgement), when the current lane ends withinspeed x lane_end_rescue_time_horizon(merge / lane drop), or when the current lanelet has no lane-follow candidate. Suppression and restoration are decided per current lanelet;enable: falsereproduces the previous behavior exactly. Pedestrians, bicycles and unknown objects are unchanged.Differences from upstream (port only, no behavior change):
declare_lane_change_path_suppression_paramsand the test fixture takerclcpp::Nodebecause this branch does not useautoware::agnocast_wrapper::Nodeinmap_based_predictionyet.declare_object_deceleration_paramshelper of upstream is not part of this branch, so only the suppression helper is added.Related links
Parent Issue:
How was this PR tested?
colcon build+colcon testofautoware_map_based_predictionon this branch in the pilot-auto.x2 (Humble) container: 182 tests, 0 failures (25 suppression tests included).TrackedObjectsthrough the modified node and unmodifiedcollision_check_filter: HIGH/FATAL segments 26 -> 12, no new segments; predicted paths per vehicle 1.98 -> 1.31; median latency 10-20 ms.Notes for reviewers
Merge together with the autoware_launch.x2 companion PR: the node declares
lane_change_path_suppression.enable/keep_when_lane_endswithout defaults, so the launch parameter file must contain the keys.Interface changes
ROS Parameter Changes
Additions and removals
lane_change_path_suppression.enablebooltruefalserestores the previous behaviorlane_change_path_suppression.keep_when_lane_endsbooltruelane_change_path_suppression.lane_end_rescue_time_horizondouble5.0speed x horizon, speed floored at 1.0 m/s)Effects on system behavior
collision_check_filter(and any consumer of all predicted paths, e.g.behavior_pathwithuse_all_predicted_paths) no longer reacts to them.lat_diff_distance, about 1 s history) no LC path is predicted; RSS /obstacle_stopare unchanged.🤖 Generated with Claude Code