Teaching agents to navigate in a map-less environment

Authors:
Shivam Goel shivam.goel@wsu.edu
Abhijay abhijay@pdx.edu

Based on:

• erlebot/gym-gazebo
• Tf Pose Estimation

Table of Contents

1. Getting Started
   a. Docker link
   b. Requirements
2. Running the experiments
   a. Run docker
   b. Clear all previously running processes
   c. Experiment
3. Visual on Gazebo
4. Useful Docker Commands

Getting Started

Docker link

Wil be updated soon

Requirements
Host machine: Ubuntu 16, CUDA Version: 10.1

Running the experiments

Run the docker

sudo docker run --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=all -it gym-gazebo_v6.1

Clear all previously running processes

ps -ef | grep ros | awk '{print $2}' | xargs kill -9
ps -ef | grep python | awk '{print $2}' | xargs kill -9
ps -ef | grep xvfb | awk '{print $2}' | xargs kill -9

Experiment

# Install conda
conda activate tf-gpu-1-5

# For entering docker  
# Open two different terminals and run  
sudo docker exec -it ef6dd4c9b971 bash  
# One is for running pose estimation and the other for dqn

# On terminal 1
cd /usr/local/gym/gym-gazebo/examples/scripts_turtlebot
run -xvfb
nohup python -u smarthome_turtlebot_lidar_dqn_withAttn_v3.py > results_withAtt_Feb16.out &

# On terminal 2 (simultaneously)
cd /usr/local/gym/tf-pose-estimation#
nohup python run_with_ros.py --model=mobilenet_thin > pose.out &

# Monitor results on terminal 1
tail -f results_withAtt_Feb16.out

Visual on Gazebo

To view the robot acting in the environment using gzserver do the following

export GAZEBO_ID=`sudo docker ps | grep gym-gazebo | awk '{print $1}'`
export GAZEBO_MASTER_IP=$(sudo docker inspect -f '{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' $GAZEBO_ID)
export GAZEBO_MASTER_URI=$GAZEBO_MASTER_IP:11345

# Note: With sudo (just copy and make sure the grep is able to get the name of the current docker)  

export GAZEBO_ID=`sudo docker ps | grep gym_gazebo | awk '{print $1}'`

gzclient

Useful Docker Commands

Check running containers

sudo docker ps -a

Start a running docker

sudo docker start <container-id>

sudo docker attach <container-id>

Creating volume

docker volume create erlerobot_test_v1
sudo docker volume ls
docker volume rm my-vol

Commit docker changes

sudo docker commit <container-id> gym-gazebo_v1.1

Create a compressed docker image to transfer to another system

sudo docker save gym-gazebo_v1.1 > gym-gazebo_v1.1.tar.gz

Load a docker image

sudo docker image load --input gym_gazebo_v1.2.0.tar.gz

Get files from the docker to hard disk

sudo docker cp c6dbab3a187a:/data/turtlebotExperiments_v1/. turtlebotExperiments_v1/.

Vice Versa:

sudo docker cp models.zip 0cccde0d68ff:/usr/local/gym/

Removing docker

docker rm <container-id>