This repository is part of the GRADE project
ROS based collision checker service based on an edited MoveIt core package.
The idea is based on the work made by @pradeepr-roboticist in this issue.
The moveit_core
is updated for ROS noetic
as of 29 Apr 2022.
My suggestion is to install moveit_core
first from apt
, and then override the package by installing this in your catkin_ws
.
The main edited file in moveit_core
is moveit_core\collision_detection_fcl\collision_env_fcl.cpp
line 333.
This function accept in input two lists of objects, checks if the objects are in the moveit planning scene (i.e. they exists), and perform a low level collision checking request (standard moveit library).
It requires as output the distance, the contacts and the cost returning this information (lines 361-367 for those settings).
The function can be called with planning_scene_->getCollisionEnv()->checkCollisionBetweenObjectGroups(ob_l1, ob_l2);
.
The object lists are simply the ids of the objects (strings) in the scene.
The collision checking node I developed than expose a ROS service that make use of this function.
The main purpose of the service is to PLACE objects in a given scene. It does that by random location guessing within a given perimeter and a given world.
You need to provide the stl/dae for all the objects you're going to use.
It also implements a checking procedure such that the object is placed within an ARBITRARY polygon.
There are two ways of using this service:
- single (always the same) object placement in the world: simply check collisions between the two meshes (one is in ob_l1, one is in ob_l2). The mesh of this object gets pre-loaded when the node is launched.
- multiple object placement in the world: recursevly check collisions between one mesh and the already placed ones. First iteration is
mesh1-world
, second ismesh2-[world,mesh1]
, etc. NOTE: note that this can be used also as a single-object placement with different meshes per each iteration.
The object movement in the scene is done in low-level moveit-based c library calls. This means that the checking can be done with a high-level of efficiency.
Simply build the node.
Run roslaunch collision_check.launch
.
This will launch:
- the collision check node
- a moveit demo scene with a
firefly
robot (generated withmoveit_setup_assistant
, based on firefly_rotors_simulator)
The firefly
will be used for "camera" placement, i.e. the single object placement. In the launch file you can edit the object
The service is listening at /fake/collision_checker/check
and is defined as follows:
string env_stl_path # ALWAYS REQUIRED
string[] ob_names # OPTIONAL
string[] ob_stl_paths # REQUIRED IF OB_NAMES SET (can be also dae)
bool is_cam
float64 forced_z # REQUIRED -1 if NOT used, otherwise set
float64[] min_limits # REQUIRED min and max limits for the rng
float64[] max_limits
geometry_msgs/Point[] env_polygon # REQUIRED the polygon to check the in/out
int32 limit_collision # REQUIRED maximum amount of collision point detected between the meshes
bool reset # reset the environment
---
float64[] x
float64[] y
float64[] z
float64[] yaw
The rule is that either is_cam
needs to be set to TRUE
OR ob_names
needs to be non-empty.
is_cam
takes preference to ob_names
(i.e. the program check is_cam
value before proceding).
is_cam
leads to option 1 above (one mesh checked against the environment, mesh pre-loaded).
ob_names
(1 or more) leads to option 2 above (different meshes checked against the environment).
As I said above is_cam
use the pre-loaded mesh (setup in the launch file as param).
When option 2 is sought ob_stl_paths
needs to have one entry for EACH object even when repeated with the full absolute path of the mesh.
You need to set min
and max
limits for x,y,z
, the polygon
vertices (x,y
). Those will be used by the uniform random number generator.
The service return the in-order x,y,z,yaw
location found. (yaw spanning the whole circle).
IF IT WAS NOT POSSIBLE TO PLACE THE OBJECT THE SYSTEM USE A FIXED VALUE
x = max_limits.at(0) 10;
y = max_limits.at(1) 10;
z = -10;
yaw = 0;
(lines 155-158 of the code).
The workflow is as follow:
- reset the scene if required or if the env changes
- create the env if necessary (red color, ID "world")
- create the object mesh in the env
- until a valid location is found generate random x,y,z,yaw values
- check for collision
- repeat a maximum of 90 times or until the maximum amount of possible collision is satisfied
- The system to work requires a moveit scene set up and running.
- By our testing this can consume a lot of RAM. Please be aware of this. We are planning to implement the same strategy with the
trimesh
package. Any contribution in that sense would be greatly appreciated.
If you find this work useful please cite our work as
@misc{https://doi.org/10.48550/arxiv.2303.04466,
doi = {10.48550/ARXIV.2303.04466},
url = {https://arxiv.org/abs/2303.04466},
author = {Bonetto, Elia and Xu, Chenghao and Ahmad, Aamir},
keywords = {Robotics (cs.RO), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {GRADE: Generating Realistic Animated Dynamic Environments for Robotics Research},
publisher = {arXiv},
year = {2023},
copyright = {arXiv.org perpetual, non-exclusive license}
}