feat: FASTLIO2 support with arm64 and hardware-verified localization

[baishibona]

Summary

• Add FASTLIO2 as alternative SLAM method alongside arise_slam (select at runtime via LOCALIZATION_METHOD)
• Add arm64 build support (open3d from source, or-tools from source, rviz2 install)
• Add bagfile mode and MAP_PATH support for localization with pre-built PCD maps
• Add ros-joy runtime dependency
• Docker multi-stage build with both arise_slam and FASTLIO2 included

Test plan

• Hardware tested on NucBox M2 Pro with Mid-360 lidar (Jazzy)
• arise_slam still works as default SLAM method
• FASTLIO2 mapping mode (PGO) verified
• FASTLIO2 localization mode with MAP_PATH verified
• Bagfile playback with use_sim_time verified
• amd64 build verified

Linear Issue

closes DIM-459

[greptile-apps]

Greptile Overview

Greptile Summary

This PR adds comprehensive FASTLIO2 support alongside the existing arise_slam, enabling runtime SLAM method selection via LOCALIZATION_METHOD environment variable. It also introduces arm64 architecture support for the Docker build.

Major Changes:

• Multi-stage Dockerfile now builds for both amd64 (desktop-full) and arm64 (ros-base) platforms
• On arm64, open3d and or-tools are built from source since pre-built binaries are x86_64 only
• Both arise_slam and FASTLIO2 configs are copied to the runtime image
• New bagfile playback mode added with use_sim_time=true for testing with recorded data
• CLI updated with --localization and --image flags for better UX
• MAP_PATH support added for localization mode with pre-built PCD maps
• ros-joy added as runtime dependency
• Build script now clones from dimensionalOS/ros-navigation-autonomy-stack fastlio2 branch (switched back from VectorRobotics fork per PR thread feedback)

Key Implementation Details:

• SLAM method selection is handled at container launch time via conditional logic in docker-compose.yml command sections
• Launch files receive use_fastlio2:=true parameter when FASTLIO2 is selected
• Hardware-tested on NucBox M2 Pro with Mid-360 lidar according to test plan
• Image size remains ~24 GB as documented

Confidence Score: 4/5

• Safe to merge with minor style improvements possible
• The PR is hardware-verified and addresses previous review feedback about switching from VectorRobotics to dimensionalOS repository. The implementation is solid with proper multi-arch support, though there are minor style improvements suggested around error handling (wildcard rm patterns, stash recovery messaging)
• docker/navigation/Dockerfile and docker/navigation/build.sh have minor style improvement suggestions but no critical issues

Important Files Changed

Filename	Overview
docker/navigation/build.sh	Updated to clone from dimensionalOS/fastlio2 branch, improved local changes detection, better error messages
docker/navigation/Dockerfile	Added arm64 support, FASTLIO2 alongside arise_slam, builds open3d and or-tools from source on arm64
docker/navigation/docker-compose.yml	Added LOCALIZATION_METHOD env var, bagfile profile, MAP_PATH support, conditional launch logic for both SLAM methods
docker/navigation/start.sh	Added --bagfile mode, --image and --localization flags, improved image existence check, creates required directories

Sequence Diagram

sequenceDiagram
    participant User
    participant build.sh
    participant start.sh
    participant Docker
    participant Container
    participant ROS2

    Note over User,ROS2: Build Phase
    User->>build.sh: ./build.sh --jazzy
    build.sh->>build.sh: Clone/update ros-navigation-autonomy-stack (fastlio2 branch)
    build.sh->>Docker: docker compose build
    Docker->>Docker: Build multi-arch image (amd64/arm64)
    alt arm64
        Docker->>Docker: Build open3d from source
        Docker->>Docker: Build or-tools from source
    end
    Docker->>Docker: Build arise_slam + FASTLIO2
    Docker->>Docker: Copy configs for both SLAM methods
    Docker-->>build.sh: Image: dimos_autonomy_stack:jazzy

    Note over User,ROS2: Runtime Phase (Hardware Mode)
    User->>start.sh: ./start.sh --hardware --localization fastlio --route-planner
    start.sh->>start.sh: Set LOCALIZATION_METHOD=fastlio
    start.sh->>start.sh: Validate .env file exists
    start.sh->>start.sh: Check lidar configuration
    start.sh->>start.sh: Verify image exists
    start.sh->>Docker: docker compose --profile hardware up
    Docker->>Container: Start with LOCALIZATION_METHOD env
    Container->>Container: source ROS workspace
    alt LOCALIZATION_METHOD=fastlio
        Container->>ROS2: ros2 launch system_real_robot_with_route_planner.launch.py use_fastlio2:=true
        ROS2->>ROS2: Start FASTLIO2 SLAM
    else LOCALIZATION_METHOD=arise_slam
        Container->>ROS2: ros2 launch system_real_robot_with_route_planner.launch.py
        ROS2->>ROS2: Start arise_slam
    end
    Container->>ROS2: Start route planner (FAR planner)
    alt USE_RVIZ=true
        Container->>ROS2: ros2 run rviz2 rviz2
    end
    Container->>ROS2: Start Foxglove bridge
    ROS2-->>User: Navigation system ready

    Note over User,ROS2: Bagfile Playback Mode
    User->>start.sh: ./start.sh --bagfile --localization fastlio
    start.sh->>Docker: docker compose --profile bagfile up
    Docker->>Container: Start with use_sim_time=true
    Container->>ROS2: ros2 launch system_bagfile.launch.py use_fastlio2:=true
    User->>Container: ros2 bag play --clock bagfile
    ROS2->>ROS2: Process bagfile with FASTLIO2

Loading

[greptile-apps]

_{4 files reviewed, 6 comments}

_{Edit Code Review Agent Settings | Greptile}

[baishibona]

DO NOT MERGE UNTIL VERIFIED ON G1 (JETSON)

[baishibona]

@greptile, @greptile-app. Update to address comments. Please review.

[greptile-apps]

_{4 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[spomichter]

@ruthwikdasyam TODO test this on the G1 onboard the jetson NX. @SUMMERxYANG add to linear. is there some way to tag /linear in github comments like this and make issues? would be super helpful