[Tracking] Testability refactoring + unit-test coverage + performance + autoware_utils consolidation across Core packages

[youtalk]

Checklist

• I've read the contribution guidelines.
• I've searched other issues and no duplicate issues were found.
• I've agreed with the maintainers that I can plan this task.

Description

This is an umbrella / tracking issue for an ongoing campaign to raise the engineering quality of autoware_core packages along four axes, one ROS package per pull request:

1. Testability refactoring — extract pure, ROS-free logic out of node methods / anonymous namespaces behind additive dependency-injection and logging seams, so core algorithms can be unit-tested without spinning up a ROS graph.
2. Unit-test coverage — add fast, deterministic unit tests (known-answer cases, edge/error branches, previously-untested helpers) and re-enable orphaned tests.
3. Performance — apply benchmark-proven, output-identical memory/compute optimizations (e.g. reserve, sliding-window, binary search, allocation removal).
4. De-duplication into autoware_utils — consolidate near-duplicate utility code into the canonical autoware_utils modules (moderate scope) to keep the Core codebase clean.

A hard constraint throughout: the public API/ABI stays additive-only (new ctors/overloads/free functions; [[deprecated]] shims for anything replaced; no struct-layout or extern template signature changes).

Purpose

• Make Core's foundational algorithm packages unit-testable in isolation, replacing slow/flaky integration-only coverage with fast deterministic tests.
• Increase test coverage measurably, package by package, under the existing Codecov gate.
• Deliver measured performance improvements (every perf change ships with a micro-benchmark and an output-equivalence guard — no regressions).
• Reduce duplicated utility code by consolidating it into autoware_utils, improving maintainability across the stack.
• Do all of the above without breaking downstream consumers (additive-only API/ABI, validated with --packages-above).

Possible approaches

Every package follows the same repeatable 6-stage playbook, shipped as a single per-package PR (plus a prerequisite PR on autoware_utils when a symbol is promoted there):

0. Characterize — capture the current green baseline (colcon test + downstream build with --packages-above); add characterization tests pinning current behavior of the code about to move.
1. Seam (additive testability refactor) — add a DI ctor/overload taking the plain param struct (or injected clock/logger); keep the existing Node& ctor as a thin forwarding shim. Extract pure logic into named, header-declared free functions.
2. Cover — add deterministic unit tests for the newly-exposed logic; re-enable orphaned tests; verify the lcov delta locally.
3. Optimize — only benchmarkable, output-identical changes, each with a committed micro-benchmark (before/after numbers) and an equivalence guard test.
4. Dedup (moderate) — replace near-duplicates of existing autoware_utils symbols; promote genuinely-shared helpers into the right autoware_utils module via a separate PR; never push lanelet/PCL/msg-specific types into generic utils.
5. Verify & PR — full package test + --packages-above on the {humble, jazzy} containers; clang-tidy / cppcheck / pre-commit clean; open the per-package PR.

Execution is orchestrated so packages can progress in parallel, with each package validated by a dual-source build of autoware_core + autoware_utils together before submission.

Definition of done

Per-package PR (the bar every PR clears):

• Public API unchanged (additive-only); ABI-safe; deprecation shims for replacements.
• Measurable unit-test coverage increase (Codecov green; net-positive lcov delta).
• Every performance change backed by a committed micro-benchmark (before/after) + an output-equivalence guard test (no regression).
• Green CI across {humble, jazzy} × {agnocast, above}; lint clean.
• Behavior-preserving: characterization + pre-existing tests still pass.
• Any autoware_utils symbol move landed as a merged autoware_utils PR the Core PR depends on.

Campaign: every package below reaches the per-package DoD.

Scope & progress (per-package PRs)

All in-scope packages now have an open per-package PR (52 total). Each box is checked when its PR merges. Every PR clears the per-package DoD above and was verified green on the fork's build-and-test CI before submission.

api/

• autoware_adapi_adaptors — refactor(autoware_adapi_adaptors): extract pure helpers and add unit tests #1147
• autoware_default_adapi — test(autoware_default_adapi): add gtest suite for pure conversion functions #1141

common/

• autoware_agnocast_wrapper — test(autoware_agnocast_wrapper): add first gtest suite and extract pure validation helpers #1132
• autoware_geography_utils — perf(autoware_geography_utils): cache EGM2008 geoid instance across height conversions #1097
• autoware_interpolation — perf(autoware_interpolation): scalar spline-eval overloads remove per-knot vector allocations #1127
• autoware_kalman_filter — perf(autoware_kalman_filter): use Cholesky solve in update() with full guard-branch test coverage #1094
• autoware_lanelet2_utils — refactor(autoware_lanelet2_utils): reuse geometry helpers, fix int-narrowing, dedup offset-bound builders #1128
• autoware_marker_utils — perf(autoware_marker_utils): hoist arc-length recompute and bundle internal cleanups #1130
• autoware_motion_utils — perf(autoware_motion_utils): avoid per-call full-container copies in trajectory offset helpers #1099
• autoware_object_recognition_utils — refactor(autoware_object_recognition_utils): make transform.hpp testable #1129
• autoware_osqp_interface — test(autoware_osqp_interface): cover uncovered branches and extract status-message builder #1110
• autoware_point_types — test(autoware_point_types): cover memory.hpp layout helpers and mark them inline #1131
• autoware_qp_interface — fix(autoware_qp_interface): report real OSQP solver status and drop redundant vector copies #1100
• autoware_signal_processing — test(autoware_signal_processing): cover Butterworth bilinear path and extract print* string builders #1146
• autoware_trajectory — test(autoware_trajectory): add Pchip interpolator unit tests #1109
• autoware_vehicle_info_utils — test(autoware_vehicle_info_utils): cover pure paths and reuse get_or_declare_parameter #1105

control/

• autoware_command_gate — refactor(autoware_command_gate): extract ROS-free mode dispatch helper #1145
• autoware_simple_pure_pursuit — refactor(autoware_simple_pure_pursuit): extract pure-pursuit math into testable free functions #1126

localization/

• autoware_ekf_localizer — test(autoware_ekf_localizer): add EKFModule unit tests via non-ROS seam #1098
• autoware_gyro_odometer — refactor(autoware_gyro_odometer): extract pure fusion logic and fix diagnostics level #1112
• autoware_localization_util — refactor(autoware_localization_util): delegate util_func helpers to autoware_utils #1106
• autoware_ndt_scan_matcher — test(autoware_ndt_scan_matcher): unit-test covariance/oscillation math and fix no-ground hot loop #1108
• autoware_pose_initializer — fix(autoware_pose_initializer): reject stale GNSS poses and deduplicate trigger modules #1095
• autoware_stop_filter — test(autoware_stop_filter): add deterministic node integration test and dedup filter eval #1148
• autoware_twist2accel — refactor(autoware_twist2accel): extract pure AccelEstimator and add unit tests #1111

map/

• autoware_lanelet2_map_visualizer — refactor(autoware_lanelet2_map_visualizer): extract pure marker-array builder and reuse autoware_utils_visualization #1143
• autoware_map_height_fitter — perf(autoware_map_height_fitter): extract testable ground-height kernel and use a cached KdTree #1107
• autoware_map_loader — refactor(autoware_map_loader): dedup PCD cell loader and cover selected module #1139
• autoware_map_projection_loader — test(autoware_map_projection_loader): add gtest for load_info_from_yaml and load_map_projector_info #1140

perception/

• autoware_euclidean_cluster_object_detector — perf(autoware_euclidean_cluster_object_detector): eliminate hot-loop allocations in clustering #1124
• autoware_ground_filter — test(autoware_ground_filter): expose grid trig + index seams and add numeric tests #1125
• autoware_perception_objects_converter — refactor(autoware_perception_objects_converter): extract pure convert() and reuse autoware_utils_uuid #1133

planning/

• autoware_behavior_velocity_planner — refactor(autoware_behavior_velocity_planner): extract node-agnostic data-ingestion helpers #1114
• autoware_behavior_velocity_planner_common — test(autoware_behavior_velocity_planner_common): cover pure helpers and dedup formatIds #1135
• autoware_behavior_velocity_stop_line_module — refactor(autoware_behavior_velocity_stop_line_module): share stop-line state machine and stop-point logic #1155
• autoware_mission_planner — refactor(autoware_mission_planner): extract pure check_reroute_safety free function #1113
• autoware_motion_velocity_obstacle_stop_module — test(autoware_motion_velocity_obstacle_stop_module): unit-test stop-decision helpers and PathLengthBuffer #1117
• autoware_motion_velocity_planner — refactor(autoware_motion_velocity_planner): extract pure node logic into testable free functions #1136
• autoware_motion_velocity_planner_common — test(autoware_motion_velocity_planner_common): cover utils pure logic and CollisionChecker #1116
• autoware_objects_of_interest_marker_interface — test(autoware_objects_of_interest_marker_interface): add marker/coloring gtest and inject optional stamp #1144
• autoware_path_generator — perf(autoware_path_generator): make goal-connection closest-point fallback lazy and hoist path length #1118
• autoware_planning_factor_interface — test(autoware_planning_factor_interface): add gtest suite and apply move-based perf wins #1152
• autoware_planning_topic_converter — refactor(autoware_planning_topic_converter): extract testable PathPoint to TrajectoryPoint conversion #1134
• autoware_route_handler — perf(autoware_route_handler): drop duplicate routing graph build in setMap #1119
• autoware_velocity_smoother — test(autoware_velocity_smoother): cover trajectory_utils pure functions and dedup getTransVector3 #1115

sensing/

• autoware_crop_box_filter — perf(autoware_crop_box_filter): identity transform fast-path and testable param merge #1138
• autoware_downsample_filters — refactor(autoware_downsample_filters): dedup layout checks via autoware_point_types and add voxel filter tests #1137
• autoware_gnss_poser — refactor(autoware_gnss_poser): drop dead helpers, per-instance prev_position, unit-test pure statics #1154
• autoware_vehicle_velocity_converter — refactor(autoware_vehicle_velocity_converter): extract pure convert() and use XYZRPY_COV_IDX #1151

testing/

• autoware_planning_test_manager — perf(autoware_planning_test_manager): harden planning_test_manager_utils #1150
• autoware_pyplot — fix(autoware_pyplot): implement declared-but-undefined Axes/Text forwarders and fix move ctors #1153
• autoware_test_utils — test(autoware_test_utils): cover pure factory functions and parser nullopt branches #1149

Related issues

• Refactoring Plan for autoware_path_generator #236 Refactoring Plan for autoware_path_generator
• Improve the package quality of copied scan_ground_filter (from autoware.universe) package #321 Improve the package quality of copied scan_ground_filter
• improve behavior_velocity_planner to Core level #322 improve behavior_velocity_planner to Core level
• Unify test map location & remove hardcoded coordinates in unit tests #665 Unify test map location & remove hardcoded coordinates in unit tests
• Refine overuse of mutable member variables #712 Refine overuse of mutable member variables