CS4278/CS5478 Assignment 3: Planning

This is the assignment 3 of CS4278/CS5478, Intelligent Robots: Algorithms and Systems, Spring 2020. In this assignment, you will be working on planning algorithms for controlling the robot to navigate on a 2D map. The decription to the assignment can be found here.

In this README, we provide information about how to run your code and generate output for submission.

Setup the environment

We developed the simulator using ROS Kinetic. The simulator is adjusted from ROS TurtleBot Stage package. As your first step, you should install the ROS kinetic according to assignment 1. After installation, install TurtleBot Stage with

sudo apt install ros-kinetic-turtlebot-stage
sudo apt install ros-kinetic-joint-state-publisher-gui

Next, clone our repo and setup your environment with

catkin_make
source devel/setup.bash

Please check your setup by

roslaunch planner turtlebot_in_stage.launch

You should be able to see the RViz and ROS stage.

What you need to do

We split the robot navigation task into three parts: simulator, controller and planner. We use a very simple controller defined in the planner file. In this assignment, we provide you with the simulator and all the necessary functions for 1) ROS nodes 2) data communication 3) controlling the robot. You only need to implement the planning algorithms and collision checking mechanism, and fill in the template marked with # TODO: FILL ME! in base_planner.py.

We would recomment you to use base_planner.py as a base class, then impelement your planners as derived classes.

How to run the code

You can simply use the following instructions to launch the simulator and set the configurations.

roscd planner/src
sh run.sh [your map name] start_x start_y start_theta
python your_planner.py --goal 'goal_x,goal_y' --com use_com_1_map_or_not

Specifically, suppose we want to load maze8.png, set the start pose of the robot as (x=2, y=2, \theta=0), and set the goal to be (x'=8, y'=8). You should:

1. Go into the source directory roscd planner/src
2. Launch the simulator and set the start of the robot sh run.sh maze8.png 2 2 0
3. Open a new shell and launch the controller python simple_controller.py
4. Open a new shell and launch the planner script with goal specified python your_planner.py --goal '8,8' --com 0. Here, --com 0 flag indicates that we are not using the com1.jpg map. This is because the environment parameter changes from other mazes to the com1 map.

Notes

For visualization, please use ROS stage. RViz provides the 2.5D visualization but has certain noise due to ROS asynchronous communication.

Submission

We provide you with 10 maps, including 9 handcrafted maps (maze0.png to maze0.png) and an illustrative COM1 level 1 floorplan (com1.jpg). You can find them here. Each map has a list of corresponding testcases, with the goals specified here. For all cases, we assume the robot starts with pose (1, 1, 0).

You should implement the planners, test them, generate controls for each testcase, and submit all of them. For discrete and continuous actions (task 1 and task 2), please save them in .txt files. For MDP policy, please save it into a json file. We have provided functions in base_planner.py.

The naming should follow {task}_{map}_{goal}.txt for task 1 and task 2; {task}_{map}_{goal_x}_{goal_y}.json for task 3. For example,

• 1_maze2_5_5.txt for the discrete planner on maze2.png with [5, 5] as the goal.
• 2_maze3_9_9.txt for the continuous planner on maze3.png with [9, 9] as the goal.
• 3_com1_43_10.json for the mdp policy on com1.jpg with [43, 10] as the goal.

Some example control files can be found here.