This is the repository of the code used for RAIM project
Paper: "RAIM: Reinforced Autonomous Intersection Management - AIM based on MADRL"
Conference: Real-World RL Workshop. 34th NeurIPS 2020 Conf
Take a look to requeriments.txt
To install a requeriments.txt file: create a new virtual environment
conda create -n RAIM python=3.8 anacondaconda activate RAIM
conda install --file requirements.txJust run the main_1_1_v2.py file
By default is used the fixed traffic light algorithm with a green time defined by: greentime=(120-10)//2
If you want to change by other traditional traffic lights algorithms, you need to instantiate in the previous lane.
Like: algorithm = REDVDAlgorithm(...)
Changing Fixed by algorithm
If you want to use the proposed algorithm, you need to instantiate the module located in TD3PER
algorithm = TD3Agent.Agent(...)
In this repository there is the code to run the paper "RAIM: Reinforced Autonomous Intersection Management - AIM based on MADRL"
In this paper, I make use of Deep Reinforcement Learning to train a new Autonomous Intersection Management (AIM) system.
AIM is a decentralyzed system located virtually in the mobile communication system that control connected autonomous vehicles at urban intersections.
Traditional AIM
Reinforced AIM, or RAIM, is an advanced technique that makes use Deep Reinforcement Learning to determine for each vehicle within an intersection or in the approaches, the speed at which it must travel during the next time interval in order to avoid collisions and minimize travel time.
RAIM makes use of Twin Delayed Deep Deterministic Policy Gradients (TD3), PER (Prioritized Experience Replay), and Curriculum-based learning through Self-Play.
RAIM actor
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update tests as appropriate.