Run

Note: All ros2 launch commands that are run on the Host PC can be run within a container of the provided docker image. See here

Full Pipeline On Host

In this particular case, there is no need to build nor deploy the ROS2 overlay for a particular edge target.

After building the ROS2 OpenVSLAM Acceleration overlay:

source $ROS2_HOME/setup.zsh 
source install/setup.zsh

ros2 launch slam_launch slam_full.launch.py dataset_path:=<dataset-path> dataset_period:=100 dataset_type:=euroc system_type:=stereo start_pangolin_viewer:=true start_rviz2:=true

By default, the first sequence of the EuRoC MAV dataset will be downloaded and which can be found in build/data_euroc/Machine_Hall_01/ relatively to the workspace.

On Target

Deployment

The deployment consists of the copying the ROS2 overlay to the board. Which, for KV260 target can be done using:

scp -r install-kv260-soallak kv260:

In addition, EuRoC MAV sequences needs to be available on the target board. One approach is to copy a dataset sequence to a USB Drive. For example to copy the sequence downloaded by default:

cp -r build/data_euroc/Machine_Hall_01 <usb-mount-point>/

Finally, plug-in the USB drive and login using e.g. ssh kv260. The last steps, on the KV260 target board, are:

1. Mount USB drive
2. Install firmware:

    cp -r install-kv260-soallak/lib/hwcv_kernels/ /lib/firmware/xilinx/
   

3. Program FPGA:

    xlnx-config -x loadapp hwcv_kernels
   

4. Test accelerated kernels:

    source /opt/ros/foxy/setup.bash
    source install-kv260-soallak/local_setup.bash 
    /install-kv260-soallak/lib/hwcv/test_all
   

Partial Pipeline

In this setup, only the accelerated stereo matching is computed on the target board. Everything else is runs on the Host PC.

On host:

source $ROS2_HOME/setup.zsh 
source install/setup.zsh
ros2 launch slam_launch slam_host.launch.py dataset_path:=build/data_euroc/Machine_Hall_01/ dataset_period:=500 dataset_type:=euroc start_rviz2:=true start_pangolin_viewer:=true

On KV260 Target:

source /opt/ros/foxy/setup.bash
source install-kv260-soallak/local_setup.bash 
ros2 launch slam_launch slam_edge.launch.py stereo_algorithm:=2

Full Pipeline

In this setup, image acquisition, rectification and accelerated stereo matching is done on the target KV260 board, while the Host PC does the poses estimation and visualization.

On Host:

source $ROS2_HOME/setup.zsh 
source install/setup.zsh
ros2 launch slam_launch slam_full_host.launch.py system_type:=stereo-depth start_rviz2:=true start_pangolin_viewer:=true

On Target:

source /opt/ros/foxy/setup.bash
source install-kv260-soallak/local_setup.bash 
ros2 launch slam_launch slam_full_edge.launch.py stereo_algorithm:=2 dataset_path:=<dataset-path> dataset_period:=200 dataset_type:=euroc

Tracing

The pipeline is instrumented using LTTng. Following are the steps to obtain the traces.

Before launching the pipeline

lttng create <session-name>
lttng enable-event --userspace slam_tracepoint_provider:'*' 
lttng start

After pipeline has finished

lttng stop

Analyzing the traces

By default, LTTng will store the traces under ~/lttng-traces/<session-name>-<date>. A script is provided to analyze LTTng traces, to use it:

ros2 run slam_tracepoint_analysis process <lttng-traces-path> <outputfile_1>.png <outputfile_2>.png