UAV Path Prediction (Guideline)

Introduction

• We use data from UAV movements in the x, y, and z axes as input to train a model using the Long Short-Term Memory (LSTM) architecture and the OSC AIMLFW framework. This trained model can predict the path of UAVs, enabling various applications.

• Ensure the Influx database is operational for this use case, which is based on the OSC AI/ML Framework (Release-J).

Suggested Requirements

Below are the suggested requirements for installing the AIMLFW:

1. OS: Ubuntu 22.04 server

2. 16 cpu cores

3. 32 GB RAM

4. 200 GB harddisk

Getting Started (Start from data insertion)

Step 1. Query influx token

Step 2. Create “UAVData” bucket (Inside Influx DB container)

Step 3. Fill the file config (UAV_insert.py)

Step 4. Excute the insert processing to insert data into Influx DB

Step 5.  Upload UAV_pipeline.ipynb to aiml-notebook and generate a UAV_pipeline.yaml

Open

Step 6. Start a UAV model training job

1. Create Feature Group: Below are some example values to be used for the standalone influx DB creation for UAV usecase.

2. Create Training Job: Below are some example values to be used for the UAV usecase training job creation when model management service is not used.

Open

Step 7. Load UAV model

Step 8. Start model prediction
1. obtain Ingress port for Kserve

2. source uav.sh

File List

• UAV_dataset.csv (Download)
  The file contains collected UAV movement path data.

• UAV_insert.py (Download)
  The file processes the UAV_dataset and inserts the data into InfluxDB.
  (Changed required: DATASET_PATH , INFLUX_IP , INFLUX_TOKEN)

• 
  An updated UAV_insert.py to write data to InfluxDB in 1000-row chunks.

• UAV_pipeline.ipynb (Download)
  The file defines the model structure and training process.

• 
  An updated UAV_pipeline to support KFP 2.2.0. Thanks to Francesco Davide Lapenta.

• UAV_deploy.yaml (Download)
  The yaml file is used for deploying model inference service.

• UAV_input.json (Download)
  The json file is the sample data for the prediction.

• UAV_predict.sh (Download)
  The script used for excuting the model prediction.

Example

Input:
This input data represents a set of three-dimensional points, where each point is defined by its normalized x, y, and z coordinates.

Example input

UAV_input

Example output

The output should be next xyz-axis path prediction (after normalization).

UAV_output

Additional Resources

List of additional documentation, links, or resources that may be helpful.

• Documentation: https://docs.o-ran-sc.org/projects/o-ran-sc-aiml-fw-aimlfw-dep/en/latest/installation-guide.html#installation-guide

• Support: Slack #AIMLFW

Acknowledge

• Joseph Thaliath, Samsung - a initial pipeline format to support AIMLFW

• Francesco Davide Lapenta, Ericsson Software Technology - updated UAV_pipeline to support KFP 2.2.0

Version History