Get instantaneous cone positions from Rtabmap pointcloud

Autonomous-Motorsports-Purdue / AMP_ASSv2

Reassembled, simpler, without move_base and loop closure issues

5 stars 2 forks source link

Get instantaneous cone positions from Rtabmap pointcloud #38

Closed WDobert closed 2 years ago

WDobert commented 2 years ago

Description

Use data produced by RTAB-Map in the obstacle pointcloud to produce an instantaneous list of visible cone positions in global space.

Sub-tasks

[ ] Subscribe to the /rtabmap/cloud_obstacles topic (type is PointCloud2)
[ ] Filter the pointcloud based on colors that match "traffic cone orange"
[ ] Cluster the pointcloud to produce a list of cone positions
[ ] Publish a ConeList on the topic /cone_finder/cones_found containing the detected cone positions

Resources

ConeList is a custom message type, defined as follows:

float32[] x
float32[] y

zghera commented 2 years ago

A rough idea for sub-tasks:

Read in pointcloud data from rtabmap bags
Color threshold the points for the cones and display them back in rviz
Plug in some pointcloud clustering library that takes the orange points (output of threshold) and create a 2D occupancy grid that can be used by move_base

WDobert commented 2 years ago

RTAB-Map may be better suited for this than we'd initially concluded from cone testing. The results we got were highly dependent on the parameters in use, but it should be possible to tune the parameters (#41) to yield better results. The full list can be accessed by using rosrun rtabmap_ros rtabmap --params.

Reschivon commented 2 years ago

This sounds great. Not having to touch the pointcloud will save significant time. We will also not have to manage loop closure as the map will be updated by RTABmap.

However, I'd suggest we test with a few cones on a sloped surface -- need to make sure that increasing the mapping precision doesn't cause it to classify the floor