feat(trajectory_selector): combine validator and concatenator

[zulfaqar-azmi-t4]

Description

Merges the previously separate autoware_trajectory_concatenator node and autoware_trajectory_validator library into a single TrajectorySelectorNode. The node collects candidate trajectories from multiple upstream planners, concatenates them into a single set, runs the configured validator plugins, and publishes the surviving trajectories.
Clean public wrappers are introduced for both subsystems (TrajectoryConcatenatorWrapper, TrajectoryValidatorWrapper) so each remains independently readable and testable. The old ValidationStage class is removed. Internal implementation types (TrajectoryValidator, ValidatorContext, TrajectoryValidatorReport) are moved under detail/.

Related links

Parent Issue:

• Link

How was this PR tested?

• colcon build — success
• colcon test — all tests pass, including:
  • Existing plugin unit tests (test_vehicle_constraint_filter, test_traffic_light_filter)
  • Node integration tests (FiltersTrajectoriesViaPlugin, HandlesPluginRejection)
  • New tests for missing-input early returns in take_validator_data() (missing odometry, acceleration, predicted objects)

Notes for reviewers

The trajectory_validator_node ROS node name and its associated topic namespace (~/input/trajectories) are replaced by the ROS node name trajectory_selector_node (executable: autoware_trajectory_selector_node) and two separate input topics (see Interface changes below). Downstream launch files that remap ~/input/trajectories will need to be updated to remap ~/input/trajectories_generative and/or ~/input/trajectories_backup.

Interface changes

Topic changes

Additions and removals

Change type	Topic Type	Topic Name	Message Type	Description
Added	Sub	~/input/trajectories_generative	autoware_internal_planning_msgs/msg/CandidateTrajectories	Candidate trajectories from generative planners
Added	Sub	~/input/trajectories_backup	autoware_internal_planning_msgs/msg/CandidateTrajectories	Candidate trajectories from backup planners
Added	Pub	~/output/trajectories	autoware_internal_planning_msgs/msg/CandidateTrajectories	Trajectories that passed all enforced validator plugins
Added	Pub	~/debug/processing_time_detail_ms/trajectory_selector	autoware_internal_debug_msgs/msg/ProcessingTimeTree	Per-function processing time breakdown
Removed	Sub	~/input/trajectories	autoware_internal_planning_msgs/msg/CandidateTrajectories	Replaced by the two topics above

Modifications

Version	Topic Type	Topic Name	Description
Old	All	/trajectory_validator_node/...	Node was named trajectory_validator_node
New	All	/trajectory_selector_node/...	Node renamed to trajectory_selector_node

Effects on system behavior

The node now concatenates trajectories from multiple sources before validation, rather than validating a single pre-concatenated input. Trajectories that arrive on trajectories_generative or trajectories_backup are merged into one candidate set each timer tick before being passed through the validator plugins. Behavior of the validator plugins themselves is unchanged.

[github-actions]

Thank you for contributing to the Autoware project!

🚧 If your pull request is in progress, switch it to draft mode.

Please ensure:

• You've checked our contribution guidelines.
• Your PR follows our pull request guidelines.
• All required CI checks pass before marking the PR ready for review.

[Copilot]

Pull request overview

This PR reduces planning pipeline node count by integrating the trajectory concatenation stage into the TrajectoryValidator node. The standalone concatenator ROS node is removed and replaced with a pure C++ autoware::trajectory_concatenator::TrajectoryConcatenator library that the validator drives via a fixed-rate timer.

Changes:

• Refactors the trajectory concatenator into a standalone C++ library and removes the ROS node/launch/config artifacts.
• Updates TrajectoryValidator to subscribe to generative + backup candidate trajectory inputs, buffer them, and run validation on a 30ms timer flush.
• Adjusts tests/launch/package dependencies to reflect the new in-process concatenation stage and topic interface.

Reviewed changes

Copilot reviewed 14 out of 14 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
planning/autoware_trajectory_validator/src/trajectory_validator_node.cpp	Adds concatenation buffering + 30ms timer-driven validation pipeline.
planning/autoware_trajectory_validator/include/autoware/trajectory_validator/trajectory_validator_node.hpp	Adds concatenator state, new subscriptions, and timer declarations.
planning/autoware_trajectory_validator/include/autoware/trajectory_validator/parameters.hpp	Introduces combined validator+concatenator parameter handling.
planning/autoware_trajectory_validator/launch/trajectory_validator.launch.xml	Switches input remaps to generative/backup trajectory topics.
planning/autoware_trajectory_validator/package.xml	Adds dependency on autoware_trajectory_concatenator.
planning/autoware_trajectory_validator/test/node/test_trajectory_validator_node.cpp	Updates integration test to publish on new input topic and set trajectory stamps.
planning/autoware_trajectory_concatenator/include/autoware/trajectory_concatenator/trajectory_concatenator.hpp	New pure C++ concatenator API.
planning/autoware_trajectory_concatenator/src/trajectory_concatenator.cpp	Implements buffering + time-based pruning + concatenation.
planning/autoware_trajectory_concatenator/CMakeLists.txt	Builds/install library instead of registering a component node executable.
planning/autoware_trajectory_concatenator/src/trajectory_concatenator_node.*	Removes standalone ROS node implementation.
planning/autoware_trajectory_concatenator/launch/trajectory_concatenator.launch.xml	Removes concatenator launch file.
planning/autoware_trajectory_concatenator/config/concatenator.param.yaml	Removes concatenator node parameter YAML.
planning/autoware_trajectory_concatenator/src/trajectory_concatenator_structs.hpp	Removes now-unused node-only parameter struct.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[danielsanchezaran]

I dont see how the inclusion of paralelism and mutex is worth it over the removal of a node that is quite simple (concatenator), can you please illustrate the advantages of this new layout? I dont think it is worth it, especially when it is not obvious that the validator should be the one concatenating trajectories

[zulfaqar-azmi-t4]

@danielsanchezaran

I dont see how the inclusion of paralelism and mutex

The mutex is not new in this PR. It's already present in the original standalone concatenator node, and I ported that logic directly into the validator without modification.

can you please illustrate the advantages of this new layout? I dont think it is worth it

There are three things to highlight

1. (Main reason) Node count reduction. This is part of the plan to reduce the number of running nodes in the planning pipeline. There is no clear requirement for concatenator to be its own node. If it is for separation of responsibility, it is satisfied just as well by keeping it as a class inside the validator. Unless there's a specific reason for it to be a node, a library called from the validator is the simpler design.
2. The standalone concat node publishes on a fixed 100ms timer. I assume that the diffusion planner has high latency which requires a longer time buffer. The side effect, however, is that the validator's callback, which based on concat's output subscription, is also capped at ~100ms cadence. Of course we can also add timer based publisher in the validator, but that doesn't make sense if the reason is because we can't wait for concatenator's subscription. Merging the nodes makes it easier to handle this.
3. The validator node is planned to be renamed to trajectory_selector, and each existing planning-pipeline node will be converted into a library called by the selector. This PR prioritizes the node merge first; the rename and the moving validator out of the package will be done later.

[danielsanchezaran]

@danielsanchezaran

I dont see how the inclusion of paralelism and mutex

The mutex is not new in this PR. It's already present in the original standalone concatenator node, and I ported that logic directly into the validator without modification.

can you please illustrate the advantages of this new layout? I dont think it is worth it

There are three things to highlight

1. (Main reason) Node count reduction. This is part of the plan to reduce the number of running nodes in the planning pipeline. There is no clear requirement for concatenator to be its own node. If it is for separation of responsibility, it is satisfied just as well by keeping it as a class inside the validator. Unless there's a specific reason for it to be a node, a library called from the validator is the simpler design.
2. The standalone concat node publishes on a fixed 100ms timer. I assume that the diffusion planner has high latency which requires a longer time buffer. The side effect, however, is that the validator's callback, which based on concat's output subscription, is also capped at ~100ms cadence. Of course we can also add timer based publisher in the validator, but that doesn't make sense if the reason is because we can't wait for concatenator's subscription. Merging the nodes makes it easier to handle this.
3. The validator node is planned to be renamed to trajectory_selector, and each existing planning-pipeline node will be converted into a library called by the selector. This PR prioritizes the node merge first; the rename and the moving validator out of the package will be done later.

The points have been addressed in a private conversation, I am ok with this PR changes given the proposed future changes. @zulfaqar-azmi-t4 please take a look at the copilot review comments too

[danielsanchezaran]

@zulfaqar-azmi-t4 please also update the READMEs