Autoware Diffusion Planner

Overview

The Autoware Diffusion Planner is a trajectory generation module for autonomous vehicles, designed to work within the Autoware ecosystem. It leverages the Diffusion Planner model, as described in the paper “Diffusion-Based Planning for Autonomous Driving with Flexible Guidance” by Zheng et al.

This planner generates smooth, feasible, and safe trajectories by considering:

• Dynamic and static obstacles
• Vehicle kinematics
• User-defined constraints
• Lanelet2 map context
• Traffic signals and speed limits

It is implemented as a ROS 2 component node, making it easy to integrate into Autoware-based stacks. The node is aimed at working within the proposed Autoware new planning framework.

How to use

(1) Prerequisites

Make sure that the directory specified in planning/autoware_diffusion_planner/config/diffusion_planner.param.yaml points to the correct model version and contains the required model weight and parameter files.

$ ls ~/autoware_data/ml_models/diffusion_planner/v4.0/
diffusion_planner.onnx diffusion_planner.param.json

This can be downloaded by following Download artifacts.

(2) Launch the planning simulator

Pass planning_setting:=diffusion_planner to switch the planning stack from the rule-based scenario planner to the diffusion planner. This argument automatically swaps the trajectory generator, the planning validator input topic, and the diagnostics graph, so no additional launch-file edits are required.

ros2 launch autoware_launch planning_simulator.launch.xml \
  map_path:=/path/to/your/map \
  vehicle_model:=sample_vehicle \
  sensor_model:=sample_sensor_kit \
  planning_setting:=diffusion_planner

Features

• Diffusion-based trajectory generation for flexible and robust planning

  

• Integration with Lanelet2 maps for lane-level context

  

• Dynamic and static obstacle handling using perception inputs

  

  

• Traffic signal and speed limit awareness

  

• ONNX Runtime inference for fast neural network execution
• ROS 2 publishers for planned trajectories, predicted objects, and debug markers

Parameters

{{ json_to_markdown(“planning/autoware_diffusion_planner/schema/diffusion_planner.schema.json”) }}

Parameters can be set via YAML (see config/diffusion_planner.param.yaml).

Inputs

Outputs

File truncated at 100 lines see the full file

Changelog for package autoware_diffusion_planner

0.53.0 (2026-09-29)

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

• docs(autoware_diffusion_planner): fix arXiv link of reference paper (#13417)

• feat(autoware_diffusion_planner): remap unsupported objects to pedestrian (#13352)

  * feat(autoware_diffusion_planner): treat HAZARD objects as pedestrians Remap HAZARD classification to PEDESTRIAN when updating agent histories so that hazard objects are considered as planner inputs. If a hazard object has a non-BOX shape, emit a warning and replace the shape with a 0.5 m bounding box. Co-Authored-By: Claude Fable 5 <<noreply@anthropic.com>>

  • feat(autoware_diffusion_planner): gate object remapping behind a parameter and extend it to UNKNOWN
  • fix(tier4_dummy_object_rviz_plugin): emit a footprint for dummy unknown objects

  * refactor(autoware_diffusion_planner): derive the unsupported-label set from the model labels ---------Co-authored-by: Yukihito Saito <<yukky.saito@gmail.com>> Co-authored-by: Claude Fable 5 <<noreply@anthropic.com>> Co-authored-by: Max-Bin <<vborisw@gmail.com>>

• feat(diffusion_planner): ego snap to prev trajectory (#13234)

  • Implemented [ego_snap_to_prev_trajectory]{.title-ref}
  • Removed [all_close]{.title-ref}
  • Applied [pre-commit run -a]{.title-ref}
  • Merge main
  • Fixed README.md
  • Fixed parameters
  • Added [max_search_segment_count]{.title-ref}
  • Fixed to false
  • Fixed to false

  * Fixed unknown words ---------

• feat(diffusion_planner): resample neighbor agent histories onto a constant grid (motion compensation) (#12979)

  • feat: Implement agent history alignment for improved neighbor agent tracking
  • Added HistoryAlignmentParams struct to configure neighbor-agent history time-alignment.
  • Introduced propagate_motion and interpolate_yaw functions for motion state propagation and yaw interpolation.
  • Updated DiffusionPlannerParams to include object_history parameters.
  • Enhanced AgentData to support history updates with alignment parameters.
  • Implemented resampling and transformation of agent histories based on alignment parameters.
  • Added tests for agent history alignment functionality, including propagation and lifecycle management.
  • Updated parameter handling in DiffusionPlanner to support new object history settings.
  • Refactor agent history alignment to resampling
  • Replaced agent history alignment with a new agent history resampler.
  • Updated CMakeLists.txt to include new source and test files.
  • Removed agent_history_alignment.hpp and agent_history_alignment.cpp.
  • Introduced agent_history_resampler.hpp and agent_history_resampler.cpp.
  • Updated AgentData to use resampling parameters instead of alignment parameters.
  • Modified tests to reflect changes in history resampling logic.
  • feat: Remove prune grace parameter and update agent history handling for immediate erasure of disappeared agents
  • feat: Rename object_history to object_motion_resampling and update related parameters for clarity
  • feat: Remove yaw rate threshold from object motion resampling parameters and update related logic

  * retain agent history of temporal absent format fix

  • object_motion_resampling parameter to read-only
  • fix velocity handling
  • chore(autoware_diffusion_planner): schema entry, spell exceptions, include cleanup
  • add object_motion_resampling to the parameter schema
  • file-scoped cspell exceptions for CTRV/dedup/tmpl
  • prune unused includes from agent.hpp; direct includes in agent.cpp

  - comment polish per codebase comment policy Co-Authored-By: Claude Fable 5 <<noreply@anthropic.com>>

File truncated at 100 lines see the full file