ssc_exploration

ssc_exploration aims at augmenting robotic exploration thanks to the use of semantic scene completion. We provide here a ROS package with a ssc node, predicting scene occupancy from the current point of view. This package integrate itself into a exploration pipeline developped in collaboration with the autonomous systems lab of ETHZ:https://github.com/ethz-asl/mav_active_3d_planning.git. The semantic scene completion itself is based on this repository.

In order to integrate the predictions of SSC to the SSC pipeline, modifications to FIESTA, unreal_cv_ros and mav_active_3d_planning were necessary. The modified versions are available here: https://github.com/arnovaill/mav_active_3d_planning https://github.com/arnovaill/unreal_cv_ros https://github.com/arnovaill/FIESTA

Architecture

For more information about the overall system, please refer to this report.

Dependencies

It is strongly recommended to use a conda environment to use this repository. The environment used to compile and run the code is provided here.

Setup

First, setup a catkin workspace using the modified version of mav_active_3d_planning based on FIESTA, following the instructions here. This should install the forked versions of mav_active_3d_planning, FIESTA, unreal_cv_ros and their dependencies.

The semantic scene completion is based on Python 3. As ROS melodic only supports Python 2.7, a separate catkin workspace based on Python 3.6 needs to be built using the following instruction:

mkdir -p ssc_ws/src
cd ssc_ws/src
git clone --recurse-submodules https://github.com/arnovaill/ssc_exploration.git
git clone https://github.com/eric-wieser/ros_numpy.git
cd ..

wstool init
wstool set -y src/geometry2 --git https://github.com/ros/geometry2 -v 0.6.5
wstool up
rosdep install --from-paths src --ignore-src -y -r

catkin build --cmake-args -DCMAKE_BUILD_TYPE=Release -DPYTHON_EXECUTABLE='/home/arno/anaconda3/envs/ssc/bin/python3.6' -DPYTHON_INCLUDE_DIR='/home/arno/anaconda3/envs/ssc/include/python3.6m' -DPYTHON_LIBRARY='/home/arno/anaconda3/envs/ssc/lib/libpython3.6m.so'

Configuration

Before any experiment, the field of view must be set corresponding to the field of view of the SSC model. In the installation folder of UnrealEngine change the field of view in /UnrealEngine/Engine/Binaries/Linux/unrealcv.ini to 50.9678. The starting height of the drone must be changed before use, it is currently hard-coded in the ssc.py file.

Run an Experiment

In the exploration workspace (Python 2), run:

roslaunch active_3d_planning_app_reconstruction example_fiesta.launch

In the SSC workspace (Python 3), run:

rosrun ssc ssc.py