B_Cubed

UCLA X1 Robotics 2025-26 B^3 Project

Drive Test Quickstart

For the current split Jetson + Raspberry Pi headless startup procedure, use:

• B_Cubed Headless Drive Quickstart

That guide includes copy-paste commands for:

• Jetson ZED/Nav2/web console startup.
• Raspberry Pi Sense HAT, RL balance controller, ROS listener, and CAN motor startup.
• Tailscale and hotspot DDS modes.
• Live checks for /cmd_vel_manual, /cmd_vel_balanced, /jet_cmd, motor powers, and RL policy status.

1. Install ROS2 Humble https://docs.ros.org/en/humble/Installation/Ubuntu-Install-Debs.html

2. Install the ZED SDK: https://www.stereolabs.com/developers/release/5.0#82af3640d775

3. Install the ROS2 Wrapper for the ZED SDK https://www.stereolabs.com/docs/ros2

4. Enable positional tracking in the ZED wrapper configuration.

Relevant wrapper parameters:

• pos_tracking.pos_tracking_enabled: true
• pos_tracking.publish_tf: true
• pos_tracking.publish_map_tf: false while slam_toolbox mapping or localization owns map -> odom
• pos_tracking.area_memory: true
• pos_tracking.two_d_mode: true for a ground robot

5. Run the ZED positional tracking bridge node after the wrapper is running: ros2 run depth_processing zed_tracking

Useful options:

• --ros-args -p input_pose_topic:=/zed/zed_node/pose
• --ros-args -p input_pose_cov_topic:=/zed/zed_node/pose_with_covariance
• --ros-args -p input_odom_topic:=/zed/zed_node/odom
• --ros-args -p input_image_topic:=/zed/zed_node/rgb/color/rect/image/compressed
• --ros-args -p input_image_is_compressed:=true
• --ros-args -p require_odom_for_localized:=true
• --ros-args -p show_visualization_window:=true

Notes:

• zed_tracking does not estimate localization itself. The ZED ROS 2 wrapper does that work; this node republishes the wrapper outputs in simpler topics for the rest of the robot.
• When you are running slam_toolbox mapping or localization, do not let a second node publish a competing map -> odom transform at the same time.
• The wrapper pose is the camera pose in the map frame. For your BB-8 style robot, that means this is the head-mounted camera pose unless you add a separate transform/fusion step to estimate the body center.
• The node now also subscribes to the ZED color stream so it can show a live camera view with localization status, pose text, and a top-down trajectory inset.
• It publishes the recent camera trajectory on zed/path, which is useful in RViz.
• It publishes an annotated image on zed/localization_view/compressed, so you can inspect the localization overlay with rqt_image_view if the OpenCV window is inconvenient.

6. Navigation groundwork

To make the ZED localization usable for mapping and Nav2, there is now a base-frame adapter, a point-cloud-to-scan bringup launch, and a first end-to-end Nav2 stack.

Base-frame adapter:

• ros2 run depth_processing zed_base_adapter
• Converts camera-centric ZED topics into:
  • zed/base_pose
• zed/base_pose_with_covariance
• zed/base_odom
• zed/base_path
• Publishes odom -> base_link by default.
• It can also publish map -> odom, but that is off by default so it does not fight whichever node already owns the global map transform.
• The default rigid transform now assumes the ZED is mounted 0.381 m above the chassis origin.

Navigation groundwork launch:

• ros2 launch depth_processing zed_nav_bringup.launch.py
• Starts:
  • zed_base_adapter
  • pointcloud_to_laserscan to generate a 2D scan from /zed/zed_node/point_cloud/cloud_registered

Mapping pass launch:

• ros2 launch depth_processing zed_mapping_pass.launch.py
• Starts:
  • zed_base_adapter
  • pointcloud_to_laserscan
  • slam_toolbox in online async mapping mode
• This is the intended first mode for a new environment.

Localization-after-mapping launch:

• ros2 launch depth_processing zed_localization_mode.launch.py map_file_name:=/absolute/path/to/posegraph_prefix
• Starts:
  • zed_base_adapter
  • pointcloud_to_laserscan
  • slam_toolbox in localization mode using a saved pose graph
• This is the intended second mode after the robot finishes its mapping pass in that environment.

Nav2 bringup on top of an already-running SLAM or localization stack:

• ros2 launch depth_processing zed_nav2_bringup.launch.py
• Starts:
  • Nav2 planner, controller, behavior server, and BT navigator
  • twist_safety_gate
• Expects:
  • map
  • /scan
  • map -> odom -> base_link
  • zed/is_localized

One-command full SLAM + navigation launch:

• ros2 launch depth_processing zed_slam_nav.launch.py slam_mode:=mapping
• ros2 launch depth_processing zed_slam_nav.launch.py slam_mode:=localization map_file_name:=/absolute/path/to/posegraph_prefix
• In mapping mode, the occupancy map continues updating while Nav2 is running.
• In localization mode, the saved environment is reused and Nav2 runs on top of that localization layer.

Safety gating:

• Nav2 publishes motion to cmd_vel_nav.
• twist_safety_gate only forwards that to cmd_vel when localization and /scan are healthy.
• Gate status is published on nav/cmd_vel_gate_status.

Useful options:

• base_to_camera_translation:=x,y,z
• base_to_camera_rpy:=roll,pitch,yaw
• base_frame:=base_link
• publish_map_to_odom_tf:=true only for ZED-only debugging when slam_toolbox is not owning map -> odom

See these docs for more detail:

• [Navigation plan](/Users/sara/Documents/My-Documents/X1 Robotics/B_Cubed/docs/navigation_plan.md)
• [Mapping to navigation workflow](/Users/sara/Documents/My-Documents/X1 Robotics/B_Cubed/docs/README_mapping_to_navigation.md)
• [Full SLAM stack](/Users/sara/Documents/My-Documents/X1 Robotics/B_Cubed/docs/README_full_slam_stack.md)
• [Handheld mapping test](/Users/sara/Documents/My-Documents/X1 Robotics/B_Cubed/docs/README_handheld_mapping_test.md)
• [Using launch.sh](/Users/sara/Documents/My-Documents/X1 Robotics/B_Cubed/docs/README_launch_sh.md)
• [ZED adapter and launch files](/Users/sara/Documents/My-Documents/X1 Robotics/B_Cubed/docs/README_zed_navigation_components.md)

7. One-stop Jetson launcher

ros2_ws/launch.sh now runs the handheld ZED mapping workflow:

• launches the ZED wrapper
• waits for the ZED pose, odom, and cloud topics
• optionally starts MediaPipe gesture recognition
• optionally starts TensorRT person tracking
• starts zed_slam_nav.launch.py in mapping mode
• opens the web planning console and prints the save commands

The web planning console is now a standalone ROS package, separate from the ZED/depth nodes: ros2 launch nav_planning_console nav_planning_console.launch.py

Useful environment overrides before running launch.sh:

• MAP_SESSION_NAME=my_env
• MAP_OUTPUT_DIR=/home/jetson-nano-x1/Documents/B_Cubed/maps
• BASE_TO_CAMERA_TRANSLATION=x,y,z
• BASE_TO_CAMERA_RPY=roll,pitch,yaw
• ZED_PARAM_OVERRIDES='general.grab_resolution:=VGA;general.grab_frame_rate:=15;pos_tracking.pos_tracking_enabled:=true;pos_tracking.area_memory:=true;pos_tracking.two_d_mode:=true;debug.use_pub_timestamps:=true'
• START_GESTURE_RECOGNITION=true
• START_PERSON_TRACKING=true
• BUILD_PARALLEL_WORKERS=2
• LAUNCH_MAX_CORES=1
• LAUNCH_MEMORY_LIMIT_MB=3072