APEIRON post processing

This repo contains instructions and tools (mostly scripts, docker files and, python notebooks) to elaborate APEIRON data.

Overview

Run or Flight session

APEIRON is divided in multiple Runs, each run is independent and contains all perception and network measurements of a single flight session. Each run is stored in a folder that can be downloaded from here using the Download All link.

Run folder description

Each run contains different files related to different sensors/acquisition device. Here's an example folder structure:

• event.bias: bias settings employed to capture data from the event-Based camera
• event.raw.timestamp: initial UNIX timestamp used as offset for the raw events, as events timestamp is relative to the initial boot of the camera
• event.raw:* event camera recording encoded in EVT3 format
• zed2i.svo:* stereocamera recordings with UNIX timestamp
• log_*.ulg: flight log, containing all the data captured by PixHawk’s sensors, such as GPS data, accelerometer, magnetometer, control inputs for the motors etc;
• tcp-internals.log:* output of ss (socket statistics)
• **tcpdump-*.pcap: ** packets capture
• run.txt: file containing meta data of the run (currently just run name)
• exported_videos: stereo camera videos exported from SVO file
  • RGB_left.avi
  • RGB_right.avi
  • depth_left.avi
• px4_csvlogs: folder containing px4 logs from the .ulg file exported in csv format (to ease the usability)
  • *.csv: various logs

suggested path and symlink

In order to avoid problems with paths used by the scripts it is suggested to:

• Download the runs and put them in a folder named "dataset" in a path of your choice
• Create a symbolic link to that folder in the parent directory of this repository: ln -s <PATH-TO-FOLDER>/dataset ../dataset_symlink

NOTE: this dataset should be stored in a High Speed Memory support (such as an SSD) to avoid problems in the processing of the SVO files.

Docker images

A set of docker images has been provided to facilitate installation of dependencies and the execution of provided scripts in a repeatable environment.

image to handle zed2i data (.SVO files) using Stereolab ZED SDK (NVIDIA-Docker required + Check ZED SDK hardware requirements)

build

navigate in svo_handling folder and build the docker image:
./build_docker.sh

access the container

navigate in svo_handling folder and start a docker container:
./zed_docker_handler.sh

image to handle event camera data (.raw files) using OpenEB SDK

build

navigate in the event_raw_handling folder and build the docker image:
./build_docker.sh

access the container

navigate in event_raw_handling folder and start a docker container:
./openeb_docker_handler.sh.sh

docker scripts

Some scripts which exploit docker images are provided as examples in the svo_handling and event_raw_handling folders.

compute zed2i visual inertial odometry

navigate in svo_handling folder and start the script:
./auto_compute_tracking.sh
this script will compute visual inertial odometry using zed sdk tracking features for all the runs stored in the suggested path(see previous sections of this document)

Jupyter Notebooks

In the notebooks folder you can find two notebooks:

• GPS_VIO.ipynb: Shows how to use zed_pose track results and px4 logs to compare the pose estimation NOTE: it is required to compute the zed2i visual inertial odometry tracking(check previous sections of this document).
• post-processBW.ipynb: Shows how to use the px4 position logs and socket statistics data to create a heatmap of the bandwidth.