ISP2021-visual_slam
This is the github project for the F1Tenth Independent Study Projects 2021. In this project we want to develop a VISUAL SLAM algorithm that is capable creating a map based on camera data as well as localize the car based on camera data.
Software Requirements
- Linux Ubuntu (tested on versions 18.04)
- ROS Melodic.
- Python 3.69.
Hardware Requiremenrts
- Realsense D435i
- Nvidia Jetson TX2
Dependencies
- Librealsense2
- Kimera
- Segmentation Pipeline
Installation
- Clone the current repository in a new workspace.
- Install Librealsense2 on your system follow the instruction listed here.
- Install Librealsense2 on the Jetson along with the kernel updates of the latest version use the instructions listed here.
- Setup and install Kimera VIO ROS and the Kimera - Semantics
- Install the Segmentation Pipeline in your current workspace.
Running the code for 3D reconstruction of map
-
Start ROS Master for communication between nodes
roscore
-
Launch the Intel RealSense camera node with the following parameters
roslaunch realsense2_camera rs_camera.launch enable_gyro:=true enable_accel:=true enable_infra1:=true enable_infra2:=true unite_imu_method:=linear_interpolation infra_width:=848 infra_height:=480 infra_fps:=15
-
Disable the camera IR emitter
rosrun dynamic_reconfigure dynparam set /camera/stereo_module emitter_enabled 0
-
Source your workspace
source ~/devel/setup.bash
-
Launching the Kimera VIO node with the follwing parameters with Loop closure detection (use_lcd) enabled
roslaunch kimera_vio_ros kimera_vio_ros_realsense_IR.launch run_stereo_dense:=true should_use_sim_time:=false use_lcd:=true
-
Launch Kimera Semantics node
roslaunch kimera_semantics_ros kimera_metric_realsense.launch run_stereo_dense:=true online:=true register_color:=true use_sim_time:=false
-
Visualise the Reconstruction on RViz
rviz -d $(rospack find kimera_semantics_ros)/rviz/kimera_semantics_euroc.rviz