Skip to content

Latest commit

 

History

History
134 lines (107 loc) · 5.42 KB

manual.md

File metadata and controls

134 lines (107 loc) · 5.42 KB

Manual Data Collection

These steps assume you have already set up the system according to the instructions.

Lidar

Setup:

  1. WARNING: Keep the LIDAR unplugged when not in use to avoid overheating (and potential damage).

  2. Set up the terminal (in the rover/lidar directory):

    source /opt/ros/noetic/setup.bash
source catkin_ws/devel/setup.bash

  3. Start roscore, xsens_ros_mti_driver. In the lidar directory:

    screen -S roscore -dm bash -c "roscore"
screen -S imu -dm bash -c "roslaunch xsens_mti_driver xsens_mti_node.launch"

On each data collection:

  1. Plug in the LIDAR. Wait until you can hear the LIDAR spinning; then, start the LIDAR in ROS.

    screen -S lidar -dm bash -c "roslaunch data_collection lidar.launch"

  2. Start collecting data.

    rosbag record -O out.bag /imu/data /ouster/imu /ouster/points
    • Data collection will create around 3.5GB of data per minute.

  3. When finished:

    • Use ctrl+C (on rosbag record...) to kill data collection when finished.
    • Unplug the LIDAR.
    • Kill the lidar ROS process with ctrl+C or, if using screen, send ctrl+C to the screen without attaching it with the command 
      screen -S lidar -p 0 -X stuff "^C"
    • Inspect the created bag with rosbag info out.bag.

Radar

Setup:

  1. Launch mmWave Studio, and wait for the initialization script to finish.

  2. Select the radar/scripts/manual_init.lua script in the bottom dropdown, and click run on the left.

    • This configures the radar and capture card. Due to software jankiness in mmWave studio, both must be re-flashed each time mmWave studio is restarted.

On each data collection:

  1. Start the data collection script. This "steals" the radar socket from mmWave studio.

    python collect.py

  2. In mmWave studio, select radar/scripts/manual_start.lua in the bottom dropdown, and click run. The data collection script should show status messages as data is collected:

    [t=1.104s] Flushing 8192 packets.
[t=1.838s] Flushing 8192 packets.
...
    • Data collection will create around 1GB of data per minute.

  3. When finished, select radar/scripts/manual_stop.lua in the bottom dropdown, and click run.

    • This will create a temporary file at the path specified in build.py (default: rover/radar/tmp.bin) which you can delete.
    • The data collection setup should raise an error (or time out).

Data Processing

Cartographer:

  1. Copy the output Lidar and Radar data to the same folder, and name them lidar.bag and radarpackets.h5 respectively.
  2. Run cartographer: 
    roslaunch slam offline_cart_3d.launch bag_filenames:=lidar.bag
    • NOTE: you may need to manually kill the process by closing the rviz window with ctrl+c after it completes.
    • This step took ~6 minutes for a 9:15 trace.
  3. Generate point cloud / occupancy grid: 
    roslaunch slam assets_writer_cart_3d.launch \
    bag_filenames:=lidar.bag pose_graph_filename:=lidar.bag.pbstream
    • This step also creates several output map images (lidar.bag_xray_{xy, yz, xz}_all.png, lidar.bag_probability_grid.png); verify that these images look about right for the scene you've captured.
    • This step took ~24 minutes for a 9:15 trace.
  4. Output poses: 
    rosrun cartographer_ros cartographer_dev_pbstream_trajectories_to_rosbag \
    --input lidar.bag.pbstream --output pose.bag
rostopic echo -b pose.bag -p trajectory_0 > trajectory.csv

Radar Processing:

  1. Only the radarpackets.h5 and trajectory.csv files are needed for radar processing.

  2. Process radar data:

    python process.py fft -p <dataset_folder>
    • Pass <dataset_folder> as the name of the folder containing radarpackets.h5 and trajectory.csv (i.e. the dataset path).
    • This should create a radar.h5 file in the <dataset_folder>.
    • You may need to reduce --batch if your GPU does not have enough memory (default: --batch 1000000)

  3. Process trajectory:

    python process.py trajectory -p <dataset_folder>
    • This should create a trajectory.h5 file.
    • Parameters to tune:
      • --smooth -1: gaussian smoothing width (higher --smooth == more smoothing) to be applied to the velocity; disabled by default (if <0).
      • --min_speed 0.2: minimum speed threshold for dataset inclusion
      • --max_accel 2.0: maximum acceleration threshold for relocalization jitter rejection
      • accel_excl 15: number of samples to exclude on either side of each detected jitter event
      • speed_excl 5: number of samples to exclude on either side when the minimum or maximum speed is exceeded

  4. Create "ground truth" lidar map:

    python process.py map pl <dataset_folder>
    • This should create a map.npz file.

  5. (Optional) Verify time synchronization:

    python speed_report.py <dataset_folder>
    • This creates a speed_report.pdf file showing the SLAM-measured and radar-inferred speed (in m/s) over time (seconds).
    • Check that the inferred speed mostly matches the measured/processed speed. In particular, pay attention to any large offsets which may indicate a time synchronization issue.
    • Make sure that not too much of the dataset is invalid (due to speeds outside the specified window).