robosense_ac_perception

README | 中文文档

1. Introduction

robosense_ac_perception provides perception capabilities such as object detection, semantic segmentation, dense depth estimation, etc.

2. Prerequisites

This project supports both ROS1 and ROS2. It has been tested on the Noetic version of ROS1 and the Humble version of ROS2.

For usage on the x86 platform, an Nvidia GPU (e.g., 3060) is required, along with the installation of CUDA and TensorRT (recommended version 8.5.2.2).

For the ARM platform, it has been adapted to the following platforms:

• RK3588 platform based on RKNN version 2.3.0
• JETSON Orin based on Jetpack 6.2
• RDK X5 based on OS Version 3.1.1

2.1 Install ROS1 or ROS2

Please follow the official instructions to install ROS:

• ROS1 Installation

• ROS2 Installation

3. Installation & Deployment

3.1 Pulling the Code

Create a new folder or enter your existing workspace, and execute the following commands to pull the code into the workspace.

cd WORKSPACE_PATH/src
mkdir ac_studio && cd ac_studio
git clone https://github.com/RoboSense-Robotics/robosense_ac_perception.git -b main

3.2 Model Download

cd robosense_ac_perception
bash download_model.sh

3.3 Installing Dependencies

Use the rosdep tool to install the dependencies required for compiling robosense_ac_perception.

cd ac_studio
rosdep install --from-paths robosense_ac_perception --ignore-src -r -y

Please follow the instructions to compile the required robosense_msgs for ROS2. Compilation is not needed for ROS1.

3.4 Compile robosense_ac_perception

For ROS1 Environment

Execute the following commands in the workspace to compile and install robosense_ac_perception.

cd WORKSPACE_PATH
catkin build perception_msgs
catkin build robosense_ac_perception

After compilation and installation, it's recommended to refresh the workspace's bash profile to ensure component functionality.

source devel/setup.bash

For ROS2 Environment

Execute the following command in the workspace to compile and install robosense_ac_perception.

cd WORKSPACE_PATH
colcon build --symlink-install --parallel-workers 8 --packages-select perception_msgs robosense_ac_perception

After compilation and installation, it's recommended to refresh the workspace's bash profile to ensure component functionality.

source install/setup.bash

4. Running

4.1 Acquiring Data

Data can be obtained either by connecting to an Active Camera online or by playing offline data packets for testing.

4.1.1 Running the Active Camera

Refer to the documentation to run the Super Sensor node for real-time data acquisition. Follow the documentation instructions to set the corresponding environment variables in the terminal to enable zero-copy mode or non-zero-copy mode for data acquisition.

4.1.2 Offline Data Playback

Use the ros command to play back data packets, for example:

# ROS1
rosbag play BAG_PATH
# ROS2
ros2 bag play BAG_PATH

4.2 Running Nodes

4.2.1 Running Object Detection Node

Run the yolov8_node node using the ros command:

# ROS1
roslaunch modules/launch/yolov8.launch
# ROS2
ros2 launch robosense_ac_perception start_yolov8.py

4.2.2 Running Semantic Segmentation Node

Run the ppseg_node node using the ros command:

# ROS1
roslaunch modules/launch/ppseg.launch
# ROS2
ros2 launch robosense_ac_perception start_ppseg.py

4.2.3 Running Dense Depth Estimation Node

Run the promptda_node node using the ros command:

# ROS1
roslaunch modules/launch/promptda.launch
# ROS2
ros2 launch robosense_ac_perception start_promptda.py

5. FAQ

Create New Issue

6. Open Source License

The Apache License, version 2.0.