MSU, Mankato's 2023 Autonomous Snowplow

YT video demo

Metapackage including ROS packages, configurations, and launch files used on the 2022-2023 MNSU autonomous snowplow project.

Sensor Pipeline:

Sensor Descriptions:

• Trimble GPS reciever, swiftnav antenna, LPMS U3 IMU, 2 1024 ppr encoders, Intel d435i RGB-D camera

• robot_localization EKF fusing data from the 3 sensors

  • Wheel odometry stream from roboteq_diff_driver - accesses motor controller through C++ api to receive encoder tick count, current in Amps through each motor channel, and temperature of motor controller
  • IMU data stream from openzen_ros C++ driver
    • Filtered & Fused orientation from magnetometer, gyro, and accelerometer data using IMU madgwick filter
  • GPS data stream from nmea_socket_pub, parsed NMEA sentences converted to robot coordinate frame using navsat_transform

Intel D435i RGB-D camera used in RTABmap package for SLAM/mapping

• Parameters and filters located here

Image Pipeline:

• RGBD_Sync nodelet: Synchronize camera messages, throttle data to improve SLAM performance
• points_xyzrgb creates pointcloud2 data from synchronized RGB-D images, used for 3-D voxel obstacle layer
• obstacle_detection package used to differentiate ground & obstacles, used for clearing & marking in costmaps
• RTABmap_ros used to generate & store grid map from RGB-D images, will include LiDAR for mapping when functional

Buffer_waypoints package is a move_base action client that allows following of multiple waypoints; will be upgraded to state machine/behavior tree in future

• subscribes to /initialpose topic published by RVIZ, adds clicked poses from that topic to a waypoint queue & begins navigating when prompted

Waypoint_follower node is similar to buffer_waypoints but includes functionality with the react web app & a workaround to move_base not changing status when waypoint is reached

Web Interface: React App that allows teleop, video stream, and autonomous functionality

nmea_socket_pub:

• reads GPS/GNSS NMEA sentence streams over specified TCP port, parsing GGA & GST sentences to both applicable GPS fix data, as well as a published covariance matrix
  • covariance matrix updated in real-time, so the kalman filter fused data takes more or less accurate GPS data into account

OpenCV scripts used to convert PNG images of a map/blueprint into binary for use in map servers.

for more technical detail on the system and explanation on software architecture decisions see this directory

Additional Project outcomes:

• Migrated the Roboteq ROS C++ driver to ROS2:

Operation

On machine running ROS master:

roslaunch navstack_pub camera.launch
rosrun web_video_server web_video_server

On rpi running web server & website client node (on same ROS master):

rosrun waypoint_follower driver
roslaunch rosbridge_server rosbridge_server.launch
serve -s build #(in react app folder; If apache not configured & react app is built)
*(to access from other network, configure husarnet and join p2p network)