-
Guillaume Duret authoredf2816e38
PickSim
PickSim is a gazebo-based pipeline to generate synthetic data.
Getting started
Picksim can easily be downloaded and tested trought a docker container at the link : https://hub.docker.com/repository/docker/guillaume0477/picksim/general or with the command "docker pull guillaume0477/picksim:v0.1" .
To make it easy for you to get started with PickSim, here's a list of recommended steps to generate example of data of the dataset FruitBin.
Run the docker containner
To run the docker container :
docker run -it --rm --device /dev/dri:/dev/dri --user gduret --gpus all -e DISPLAY=$DISPLAY -e NVIDIA_VISIBLE_DEVICES=all -e NVIDIA_DRIVER_CAPABILITIES=all --mount type=bind,source=/home/gduret/dataset_host,target=/dataset -v /tmp/.X11-unix:/tmp/.X11-unix:rw guillaume0477/picksim:v0.1
Preprocess
The first command command is used to download assets of fruits
roslaunch dataset install.launch name:=objects version:=default
The second command is used to process the raw meshes to get sdf files for the gazebo simaultion until the statement : "[s2rg/blender-1] process has finished cleanly".
roslaunch dataset generate.launch name:=objects version:=optimized k_loop:=0
The following command is used to randomize the scene and generate at the last step world files of each scenes with all different lights. The statement "[s2rg/blender-1] process has finished cleanly" allow to be sure that everithing has done successfully.
roslaunch dataset generate.launch name:=s2rg version:=Fruits_all_medium k_loop:=0
/opt/learn-real/ros/melodic/integration/src/s2rg/config/version/generator/Fruits_all_medium.yaml can be modify to change the number of scene generated.
Recording
The record command uses the pregenerated world file to record the camera data for the scenes 1 to 5
roslaunch dataset record.launch name:=s2rg version:=Fruits_all_medium begin:="0" end:="5"
It tould take some time to generate until the program reach the stateement : [s2rg/recorder-2] process has finished cleanly
For more personalisation, /opt/learn-real/ros/melodic/integration/src/s2rg/config/recorder.yaml allow to change the moment when the record is effectued and which features have to be recorded, it implicitly control how many images are recorded in the scene.
Visualization
To visualize the dataset in rviz the command is :
roslaunch dataset read.launch name:=s2rg version:=Fruits_all_medium
Other scenes
Fruits_all_medium can be replaced to get other scene arangement as : "Fruits_apple_bin.yaml", "Fruits_apricot_bin.yaml", "Fruits_banana_bin.yaml", "Fruits_cherry_bin.yaml", "Fruits_kiwi_bin.yaml", "Fruits_lemon_bin.yaml", "Fruits_orange_bin.yaml", "Fruits_peach_bin.yaml", "Fruits_pear_bin.yaml"
Scenes with ycb objects can be generated with the commands :
roslaunch dataset install.launch name:=ycb version:=retopology
roslaunch dataset generate.launch name:=ycb version:=standard k_loop:=0
roslaunch dataset generate.launch name:=s2rg version:=ycb_retopology k_loop:=0
roslaunch dataset record.launch name:=s2rg version:=ycb_retopology begin:="0" end:="5"
Roadmap
PickSim is accessble by docker to be easier to install and test. The source files will be added as soon as possible in this repo.