List view
Implement a Gaussian distribution of frequency on artificial queues Experiments with 2TLCS QPLEX as future work ?
Due by May 21, 2021•1/1 issues closedTODO : - make the agents know the other agent action - implement artificial queue instead of dead end (make it modulable) - modify generator to make ~ 5% coming flow from sources not managed yet -> make more realistic scenarios behavior as a goal
Due by April 30, 2021•4/4 issues closed2-intersections simulation focus : - static light as a reference point - test part of code
Due by April 23, 2021•2/2 issues closed- Minor updates coming : occupancy and evaluation of metrics - Test the impact of the modification of green light duration on single intersection - Start setting up the 2 intersection simulation with 2 agents (totally independent first) Due date pushed from 9 to 16 april (takes longer than expected)
Due by April 16, 2021•5/5 issues closedSolve dummy problem : from 1 agent per scenario to 1 agent for all scenarios Test different model size to analyse the behaviors/results (early stopping necessary ?) Test if the order of scenario in replay (training) affects the results of the agent + multiprocessing python
Due by March 26, 2021•3/3 issues closedAnalyse the reinforcement learning model : - plot the loss - modify the cumulative waiting time metric - create fundamental diagram of traffic flow Find a better NN model
Due by March 17, 2021•3/3 issues closedApply some changes to the state of the agent (per each cell) Improve CNN model Play with code
Due by March 10, 2021•4/4 issues closed