Navigation package for Differential Drive robots

This is my implementation of a complete 2D navigation package, including global planner, local planner, and motion controller (path follower).

I developed the above 3 modules, corresponding to the folders:

• /global_planner
• /local_planner
• /stage_robot_controller

This folder contains the launch file for RVIZ, Gmapping, Stage_ros, Gazebo_ros:

• /launch_stage_gmapping_rviz

Required input messages:

• nav_msgs/OccupancyGrid
• nav_msgs/Odometry
• sensor_msgs/LaserScan

Output message:

• /cmd_vel, geometry_msgs/Twist

Two simulators are used for demonstration purpose:

1. stage_ros .

   • It uses a bmp image to build the world. I draw my map using this bitmap tool pixilart .
   • It creates a differential-drive robot with 2D lidar.

2. Gazebo with turtlebot .

   • The map is turtlebot_house.
   • The turtlebot robot is a differential-drive robot with 2D lidar. It also has RGBD cameras though not used in this project.

Robot exploring the map to reach the goal:

For slam, I am using the standard Gmapping for demonstration.

rqt_grpah

This is the workflow within the system (using stage_ros simulator).

ROS messages used:

• geometry_msgs: Twist, PoseStamped
• sensor_msgs: LaserScan
• nav_msgs: OccupancyGrid, MapMetaData, Odometry, Path

C++ libraries used:

• ros/ros.h
• vector, array, queue, string
• atomic, thread
• iostream, chrono, stdexcept, iomanip
• set, map, unordered_map
• limits, algorithm, math.h

Build

Place the needed folders in a ros workspace, e.g. catkin_ws/src/, and run catkin_make at catkin_ws

To-Do

• Add algos from random-tree family, and compare with the A* family here.
• Add MPC Model-predictive-controller.
• Add Reed-shepp curve feature.

---

Global goal and Global path

The line of green arrows is the path, from current location to the goal location.

This example is using A* algorithm to find the path.

Each path-finding takes <1~20 ms. The more obstacles between robot and goal, the longer time it takes to find the valid path.

But this node is set to update the global path about 2~3 Hz even though it could do much faster, since global path doesn't need to be updated too frequently.

Here I am dragging the robot in Stage simulator by hand.

---

Local goal and Local path

The green arrows are global path, and the red ones are local path.

This example is using Hybrid-A* to find the local path. The local goal is the last glocal path-point covered inside the local search area (The white square area around the robot base_link).

This is also a full working demo actually, since the robot is moving following the cmd_vel message generated by my controller module.

---

Working demo

Several new goals are added while the robot is moving. The paths are updated to guide the robot.

The controller module use odometry and local_path to generate the disired linear & angular velocity for the robot to follow the local_path.

In the planning module, the unkown region (dark green) is treated the same as clear region (bright white). Becasue the global planning is consantly updating the global path, a new path will be found if the existing path pass through a new-found obstacle.

The current path follower is Pure pursuit.

1. Stage sim

2. Gazebo

---