Planning Evaluator

Purpose

This package provides nodes that generate various metrics to evaluate the quality of planning.

Metrics can be published in real time and saved to a JSON file when the node is shut down:

• metrics_for_publish:
  • Metrics listed in metrics_for_publish are calculated and published to the topic.
• metrics_for_output:
  • Metrics listed in metrics_for_output are saved to a JSON file when the node shuts down (if output_metrics is set to true).
  • These metrics include statistics derived from metrics_for_publish and additional information such as parameters and descriptions.

Metrics

All possible metrics are defined in the Metric enumeration located in:

• include/autoware/planning_evaluator/metrics/metric.hpp
• include/autoware/planning_evaluator/metrics/output_metric.hpp

These files also provide string conversions and human-readable descriptions for use in output files.

Metric Classifications

By Data Type

1. Statistics-based Metrics:
   • Calculated using autoware_utils::Accumulator, which tracks minimum, maximum, mean, and count values.
   • Sub-metrics: /mean, /min, /max, and /count.
2. Value-based Metrics:
   • Metrics with a single value.
   • Sub-metrics: /value.
   • Some metrics with older implementations use the statistics-based format of /mean, /min, /max, but all values are the same.
3. Count-based Metrics:
   • Count occurrences of specific events over a time period.
   • Sub-metrics: /count, or /count_in_duration.

By Purpose

1. Trajectory Metrics
2. Trajectory Deviation Metrics
3. Trajectory Stability Metrics
4. Trajectory Obstacle Metrics
5. Modified Goal Metrics
6. Planning Factor Metrics
7. Steering Metrics
8. Blinker Metrics
9. Other Information

Detailed Metrics

Trajectory Metrics

Evaluates the current trajectory T(0) itself. Metrics are calculated and published when a trajectory is received.

Implemented metrics

• curvature: Statistics of curvature at each trajectory point.
  • Sub-metrics to publish: /mean, /min, /max.
  • Sub-metrics to output: The same as above but take the published data as data point instead of each trajectory point.
• point_interval: Statistics of distances between consecutive trajectory points.
  • Sub-metrics to publish: /mean, /min, /max.
  • Sub-metrics to output: The same as above but take the published data as data point instead of each trajectory point.
• relative_angle: Statistics of angles between consecutive trajectory points.
  • Parameters: trajectory.min_point_dist_m (minimum distance between points).
  • Sub-metrics to publish: /mean, /min, /max.
  • Sub-metrics to output: The same as above but take the published data as data point instead of each trajectory point.

• resampled_relative_angle: Similar to relative_angle, but considers a point at a fixed distance (e.g., half the vehicle length) for angle calculation from the current point as the next point to calculate the relative angle, instead of using the immediately adjacent point.

• length: Total trajectory length.
  • Sub-metrics: Value-based metric, but using the statistics-based format of /mean, /min, /max with the same value.
  • Sub-metrics to output: /mean, /min, /max for the published data.
• duration: Expected driving time to travel the trajectory.
  • Sub-metrics: Value-based metric, but using the statistics-based format of /mean, /min, /max with the same value.
  • Sub-metrics to output: /mean, /min, /max for the published data.
• velocity: Statistics of velocity at each trajectory point.
  • Sub-metrics to publish: /mean, /min, /max.
  • Sub-metrics to output: The same as above but take the published data as data point instead of each trajectory point.
• acceleration: Statistics of acceleration at each trajectory point.
  • Sub-metrics to publish: /mean, /min, /max.
  • Sub-metrics to output: The same as above but take the published data as data point instead of each trajectory point.
• jerk: Statistics of jerk at each trajectory point.
  • Sub-metrics to publish: /mean, /min, /max.
  • Sub-metrics to output: The same as above but take the published data as data point instead of each trajectory point.

Trajectory Deviation Metrics

Evaluates the trajectory deviation by comparing the trajectory T(0) and the reference trajectory.

File truncated at 100 lines see the full file

Changelog for package autoware_planning_evaluator

0.53.0 (2026-09-29)

• Merge remote-tracking branch 'origin/main' into prepare-0.53.0-changelog

• fix(design): align the evaluator node designs with the packages they describe (#13341) The fixed-name ports of the evaluation adapters, the online perception evaluator and the metric converter are pinned with global:, which keeps them out of the design graph: link_manager skips any connection whose target is a global input port and the exporter emits no remap. remap_target: keeps the same topic and service names while letting the ports take part in the graph. /diagnostics stays global:. ControlEvaluator names a plugin class in the wrong namespace, publishes tier4_metric_msgs/MetricArray rather than the non-existent autoware_control_msgs/ControlEvaluation, and subscribes to eleven planning factor topics under /planning/planning_factors, two of which were missing. PlanningEvaluator publishes the same metric type. The evaluation adapter nodes are components with no executable of their own.

• fix(autoware_planning_evaluator): fix out-of-bounds read in calc_lookahead_trajectory_distance (#13272)

• fix(evaluator): declare the dependencies these packages use (#13218) Six packages use headers or symbols of packages that they never declare. Add the 26 missing entries: 25 <depend> and 1 <test_depend>. A per-package sweep of the first commit found four more missing entries. autoware_kinematic_evaluator and autoware_localization_evaluator call find_package(ament_cmake_ros REQUIRED) under BUILD_TESTING and declare it nowhere. Both get <test_depend>ament_cmake_ros</test_depend>. autoware_planning_evaluator uses lanelet::ConstLanelet in the library code, and the header arrives only through autoware_lanelet2_utils. It gets <depend>lanelet2_core</depend>. Its test loads two parameter files from the share directory of autoware_test_utils, so it gets <test_depend>autoware_test_utils</test_depend>.

• refactor(evaluator): move node design files into each package (#13099) Co-authored-by: Claude Opus 5 (1M context) <<noreply@anthropic.com>>

• fix(autoware_planning_evaluator): use Newest polling policy for route/vector_map (#13055)

• refactor(autoware_planning_evaluator): migrate to polling:: API (#13035)

• feat(planning_evaluator): apply [agnocast_wrapper::Node]{.title-ref} to [planning_evaluator]{.title-ref} (#12966)

  • apply agnocast_wrapper::Node
  • fix
  • use agnocast_wrapper tf2
  • refactor(autoware_planning_evaluator): minimize diff from origin/main
  • Restore the plain '// ROS subscribers' comment

  - Drop explicit QoS on subscribers that had no explicit QoS in origin/main (keep transient_local QoS only on route/vector_map)

  • Remove the allow_same_message explanatory comments

  * refactor(autoware_planning_evaluator): drop explicit QoS on planning_factors sub origin/main created this subscriber with no explicit QoS.

  • style(pre-commit): autofix
  • fix: remove arg

  * fix: cpplint ---------Co-authored-by: pre-commit-ci-lite[bot] <117423508+pre-commit-ci-lite[bot]@users.noreply.github.com> Co-authored-by: kobayu858 <<yutaro.kobayashi.2@tier4.jp>>

• fix(planning_evaluator): guard degenerate curvature samples in trajectory metrics (#12906) Skip curvature calculation when the triangle edge-length product is near zero to avoid NaN/Inf values polluting trajectory curvature statistics. Co-authored-by: Cursor <<cursoragent@cursor.com>>

• Contributors: Koichi Imai, Mete Fatih Cırıt, Ryohsuke Mitsudome, Taekjin LEE, Yuxuan Liu

0.52.0 (2026-06-30)

• Merge remote-tracking branch 'origin/main' into tmp/bot/bump_version_base

• feat(autoware_vehicle_info_utils): refactor to use createFootprint

File truncated at 100 lines see the full file