dlonng / rspmp

RSPMP: real-time semantic perception and motion planning for autonomous navigation of unmanned ground vehicle in off-road environments

RSPMP

RSPMP: https://link.springer.com/article/10.1007/s10489-022-03283-z

Build

environment

• OS：Ubuntu16.04 or Ubuntu18.04
• ROS: kinetic or Melodic
• Depend：octomap_server, gridmap

1. git clone [email protected]:DLonng/RSPMP.git
2. catkin_make
3. according error to apt install lib：sudo apt-get install ros-[dist]-[xxx]

Node

agilex_msgs

• custom msg for semantic pointcloud

lidar_camera_fusion

• fun: fusion pointcloud and semantic img, output semantic pointcloud
• input: raw pointcloud, raw semantic img, calibration matrix, semantic confidence
• output: semantic pointcloud

octomap_generator

• fun: based on the input semantic point cloud and localization information, real-time construction of a local semantic 3D voxel map centered around the vehicle, which is then projected into a 2D cost map for subsequent local path planning
• input: fusion semantic pointcloud, location tf
• output: semantic 3D voxel map, 2D costmap

sem_dwa_planner

• fun: incorporate DWA (Dynamic Window Approach) for local planning based on the 2D costmap, with DWA introducing semantic costs. The planning process minimizes semantic costs.
• input: 2D costmap, ego status
• output: ego control signals

semantic segmentation

This project does not provide a model for the semantic segmentation node. You can use any real-time semantic segmentation model of your choice.

Run

Mapping

1. Only mapping, you can record dataset
2. dataset topics：/raw_img, /raw_lidar, /tf，/semantic_img, /semantic_confidence
3. start fusion node：roslaunch lidar_camera_fusion lidar_camera_fusion_no_msg_sem.launch
4. start mapping node：roslaunch octomap_generator octomap_generator.launch
5. fusion + mapping：roalaunch octomap_generator octomap_generator_mapping_scout.launch

Planning

• This project does not provide a simulation. I used a vehicle directly at the time.（Agilex Scout）
• 1. roalaunch sem_dwa_planner sem_dwa_planner.launch
• start nodes in vehicle：
  1. all sensor nodes
  2. fusion node：https://www.bilibili.com/video/BV1Sp4y1S74w/?spm_id_from=333.999.0.0
  3. mapping node：https://www.bilibili.com/video/BV1Q64y1D75N/?spm_id_from=333.999.0.0
  4. planning node：https://www.bilibili.com/video/BV1eT4y1P7bZ/?spm_id_from=333.999.0.0
• set a point in rviz
• 1. roalaunch sem_dwa_planner sem_dwa_planner_tf_rviz.launch

reference project

evelation_mapping

rspmp's Issues

运行介绍

请问大佬有相关链接可以看一下是怎么运行和仿真吗

置信度矩阵的提供问题

非常感谢您做出的出色工作。我使用的是不是LEDnet的语义分割框架，所以想请教一下点云与图像进行语义融合那里提供的置信度矩阵是如何获得的呢？谢谢！！

仿真

大佬，这套代码能否出个教程运用到仿真里面行吗？或者能讲解一下是怎么用的吗