ABZ 2025 Case Study

Training and testing of agents

Setup:

python -m venv env
source env/bin/activate
pip install -r requirements.txt

The agent for the Highway Environment can be trained or tested via command line:

# To train a new model (save under models/highway_env):
python highway_agent.py train


# Loads the trained model and runs and renders episodes over its behaviour:
python highway_agent.py test

Scenarios

In the following, we provide some scenarios. The controlled vehicle (ego vehicle) is marked in red.

Scenario 1

A vehicle approaches another vehicle in the same lane from behind, the distance decreases.	The rear vehicle brakes.	The rear vehicle continues braking to maintain the safety distance.

Scenario 2

A vehicle approaches another vehicle in the same lane from behind, the distance decreases.	The rear vehicle switches to the right lane.	The rear vehicle completes switching to the right lane and accelerates.

Scenario 3

There is another vehicle on another lane further right.	The rear vehicle switches to the right lane, as the safety distance is maintained.	The rear vehicle continues driving forward.

Scenario 5

There is another vehicle on another lane further right.	The rear vehicle cannot switch to the center lane, as the safety distance is not maintained.	The rear vehicle continues driving forward.