This repository contains the code for predictive quality of service (pQoS) for teleoperation of autonomous vehicle.
The method aims to predict two important metrics for teleoperation: uplink bandwidth and round-trip time latency.
For documentation of the code, please refer to documentation
The concept of the pQoS pipeline is to use historically recorder QoS data as anchor points to mitigate the concept drift of the trained machine learning models. The machine learing models are trained to predict the difference between the anchor points and the target values, instead of directly prdicting the target value.
The data is stored at mediaTUM and can be downloaded from here.
Put the downloaded data under data, before running the examples.
The repository contains two best performing models for uplink and latency prediction based on parameter searching.
Both UL_n_100_depth_4_state_42_lags_60_steps_60.pkl and Latency_n_100_depth_4_state_42_lags_60_steps_60.pkl takes 60 frames (lag) of last measurements and predicts 60 frames of future target (step). The input and target features are shown:
| Models | Input Feature | Target Feature |
|---|---|---|
| Uplink Model | Latitude, Longitude, RSRQ, RSRP, SINR, CQI | UL_difference |
| Latency Model | Latitude, Longitude, Latency, TXbitrate, RSRQ, RSRP, SINR, CQI, Latency_prediction, Latency_difference |
Latency_difference |
Create a virtual environment with venv
python3 -m venv venv
source venv/bin/activate
Install dependencies
python3 -m pip install -r requirements
Run the evaluation example with
source venv/vin/activate && python3 script/eval.py
The related paper is currently under publication process. A citation will be added here once the publication becomes available.