Planning Data Analyzer

Overview

This package provides offline evaluation tools for trajectory planning performance analysis from recorded rosbag data.

Open-Loop Evaluation

Open-loop evaluation computes both trajectory error metrics and planning-quality subscores.

Trajectory Error Metrics

For each evaluated trajectory, the following are computed:

• ADE
• FDE
• AHE
• FHE
• lateral deviation
• longitudinal deviation
• TTC trace
• per-horizon metrics (full, and configured horizons such as 1.0s, 2.0s, 4.0s, 8.0s)

Planning Quality Score (EPDMS)

To show multi-facet (comfort, progress, safety) quality of generated trajectory, EPDMS (Extended Predictive Driver Model Score) is computed based on the multiple subscores.

Detailed definitions and equations for each subscore can be found in the EPDMS Metrics Document.

Two variant topics are exported:

• synthetic_epdms_raw
• synthetic_epdms_human_filtered

EPDMS’s Subscores

EPDMS is obtained from combining the following subscores:

• history_comfort
• extended_comfort
• time_to_collision_within_bound
• lane_keeping
• ego_progress
• drivable_area_compliance
• no_at_fault_collision
• driving_direction_compliance
• traffic_light_compliance

These subscores are exported both as per-trajectory outputs and as aggregate statistics.

Evaluator Metrics (bag-recorded scalars)

It aggregates the results of filtering the topics specified in config/evaluator_config.yaml according to the exception rules.

• lateral_deviation_centerline
  • for excluded intersection_area

Quick Start

Open Loop Evaluation

ros2 run autoware_planning_data_analyzer autoware_planning_data_analyzer_node --ros-args \
  -p bag_path:=~/my_bag \
  -p evaluation.mode:=open_loop \
  -p trajectory_topic:=/planning/trajectory \
  -p json_output_path:=~/results.json

Output Files

• evaluation_result.json - Detailed metrics (ADE, FDE, TTC, etc.) and summary statistics
• evaluation_output.bag/ - Evaluation metrics as rosbag for visualization
• debug_images/*.png - Trajectory visualization (with --viz flag)
• evaluator_metric.json — evaluator metric aggregates per evaluator_config (if bag contains configured topics)

Parameters

Key parameters in config/planning_data_analyzer.param.yaml:

• evaluation.mode: Evaluation mode (open_loop)
• evaluation_interval_ms: Sampling interval (default: 100ms)
• sync_tolerance_ms: Time synchronization tolerance (default: 100ms)
• trajectory_topic: Trajectory topic to evaluate
• open_loop.enable_epdms_calculation: Enables EPDMS subscore, synthetic EPDMS, and EPDMS debug calculation
• open_loop.*: Open-loop specific settings

For More Information

• Configuration: config/planning_data_analyzer.param.yaml, config/evaluator_config.yaml
• Launch files: launch/

Changelog for package autoware_planning_data_analyzer

0.8.0 (2026-10-19)

• feat(planning_data_analyzer): expose EPDMS debug launch gate (#459)

• feat(planning_data_analyzer): add EPDMS calculation gate (#458)

  • feat(planning_data_analyzer): add EPDMS calculation gate
  • fix(planning_data_analyzer): gate EPDMS JSON output

  * fix(planning_data_analyzer): remove unreachable EPDMS branch ---------

• fix(planning_data_analyzer): flatten evaluator metric JSON output format (#448)

  • fix(planning_data_analyzer): flatten evaluator metric JSON output format

  * delete extra change ---------Co-authored-by: t4-adc <<grp-rd-1-adc-admin@tier4.jp>>

• feat(planning_data_analyzer): add EPDMS debug topics (#441)

  • feat(planning_data_analyzer): add EPDMS debug topics
  • fix(planning_data_analyzer): keep LK debug fields append-only
  • fix(planning_data_analyzer): restore source faithful EPDMS debug markers
  • fix(planning_data_analyzer): align TLC and TTC debug marker semantics
  • refactor(planning_data_analyzer): reduce EPDMS debug helper duplication
  • fix(planning_data_analyzer): use Autoware debug helper utilities
  • refactor(planning_data_analyzer): reuse debug helpers
  • fix(planning_data_analyzer): gate DDC debug context collection
  • fix(planning_data_analyzer): gate NC horizon debug collection
  • fix(planning_data_analyzer): clean EP debug summary fields
  • fix(planning_data_analyzer): split epdms debug gates

  * fix(planning_data_analyzer): gate epdms debug payload collection ---------Co-authored-by: beomseok-kimm <<beomseok.kim.2@tier4.jp>>

• docs(planning_data_analyzer): describe EPDMS output topics (#442)

  • docs(planning_data_analyzer): describe EPDMS output topics

  * style(pre-commit): autofix ---------Co-authored-by: beomseok-kimm <<beomseok.kim.2@tier4.jp>> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

• feat(planning_data_analyzer): filtering the [lateral_deviation_centerline]{.title-ref} by [intersection_area]{.title-ref} (#440)

  • tmp
  • fix
  • fix readme

  * addressed comments ---------Co-authored-by: t4-adc <<grp-rd-1-adc-admin@tier4.jp>>

• docs(planning_data_analyzer): add EPDMS metric equations (#438)

  • docs(planning_data_analyzer): add EPDMS metric equations
  • docs(planning_data_analyzer): address sakayori's review for EPDMS doc
  • Link epdms_metrics.md in README
  • Refine velocity notation to vector (v_t -> vec{v}_t)
  • Rename 'hard admissibility gate' to 'multiplicative penalty gate'
  • Clarify tiered penalties (0.5) for NC and DDC
  • Correct explanation of single-proposal EP simplification
  • refactor(planning_data_analyzer): added clickable functions on the epdms metric md file
  • style(pre-commit): autofix
  • refactor: Added parameter table
  • style(pre-commit): autofix
  • refactor: refined explanation

  * style(pre-commit): autofix ---------Co-authored-by: beomseok-kimm <<beomseok.kim.2@tier4.jp>> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

• feat(planning_data_analyzer): migrate EP metric (#436) Co-authored-by: beomseok-kimm <<beomseok.kim.2@tier4.jp>>

• feat(planning_data_analyzer): update EPDMS aggregation filter (#437)

File truncated at 100 lines see the full file