OpenAI gym simulator for the INTERACTION dataset
To use this simulator, first download the INTERACTION dataset (above). The dataset contains two folders which should be copied into a folder called ./datasets:
- the contents of
recorded_trackfilesshould be copied to./datasets/trackfiles - the contents of
mapsshould be copied to./datasets/maps
Install requirements and build environment with pip:
pip install -r requirements.txt
pip install -e .
To ensure correct installation, tests can be run with pytest.
If you have issues with ffmpeg and are using Conda to manage environment, you can resolve this with conda install -c conda-forge ffmpeg
The intersim-v0 gym environment loads a track and map file from the dataset. It sets the vehicle tracks and spawn times from the dataset, estimating tracks with a 20th order polynomial. The user specifies a single acceleration action for each vehicle present in the scene, and the vehicles follow double integrator dynamics along their tracks. The environment supports customizable observations, as well as an interaction graph class in order to mask the observations based on which vehicles are 'interacting'. This environment is appropriate for multi-agent decision making. For details, see intersim/envs/simulator.py.
An intersim environment can be initialized with:
import gym
env=gym.make('intersim:intersim-v0')
The intersimple environment classes build on top of intersim-v0 to allow for the control of a single agent while the remaining agents follow their estimated accelerations (and therefore recorded tracks) from the dataset. Accelerations are estimated from the positions in the datasets while using simple L2 smoother. For details, see intersim/envs/intersimple.py.
Using env.render() will add the environment states and actions to a replay buffer. env.close(filestr='output/render') will then animate the buffer and save the video, along with all relevant data to a file location with base string 'output/render_'.