Adaptive Auctions for Traffic Coordination

Repository for the Master Thesis "Adaptive Auctions for Traffic Coordination"

Abstract

The focus of this thesis lies on the relatively unexplored topic of using auctions to coordinate traffic in intersections. The primary advantage of this method over the traditional traffic light system is that it takes into account the driver’s urgency, allowing drivers in a hurry to bid more to receive priority. This thesis explores whether there is a benefit in using adaptive, instead of traditional, auctions for the coordination of traffic in a road network. A traffic simulation is built where Adaptive auctions are tasked with assigning priority by balancing the cars’ waiting time and their bids. This design is compared with Random and Zero auctions (random and no consideration of waiting time respectively). Compared to Random and Zero auctions, the current approach led to better traffic dispersion, overall lower waiting times and circumstantially higher trip satisfaction. Furthermore, it was found that there is a tendency to minimise drivers’ waiting time instead of giving priority to agents with high urgency. While these findings highlight the potential of Adaptive auctions, their practicality in a traffic environment is limited and challenging to observe with small sample sizes, despite being statistically significant.

How To Run

Requirements

The requirements for this project can be found in the requirements.txt file. To install them, run the following command in the terminal:

pip install -r requirements.txt

Running the Simulation

To run the simulation with the default parameters, simply run:

python aatc.py run

A number of parameters can be changed, as described in the table below:

Optional Parameters

Argument Name	Default Value	Choices	Type	Help Description
--num_of_epochs	10000	[1-100000]	int	Number of epochs to run. Defaults to 10000. Must be an integer between 1 and 100000.
--num_of_simulations	50	[1-10000]	int	Number of simulations to run. Defaults to 50. Must be an integer between 1 and 10000.
--grid_size	3	[2-9]	int	Size of the traffic grid. Defaults to 3 (9 intersections). Must be an integer between 1 and 9.
--queue_capacity	10	[1-100]	int	Capacity of each car queue. Defaults to 10. Must be an integer between 1 and 100.
--congestion_rate	0.07	0.01-1	float	Rate of congestion. Defaults to 0.07. Must be a float between 0.01 and 1.
--with_hotspots			Boolean	If enabled, there are periodic hotspots to increase congestion in different intersections.
--wage_time	10	[1-100]	int	Number of epochs between wage distributions. Defaults to 10. Must be an integer between 1 and 100.
--credit_balance	5	1-100000	float	Initial & Renewal credit balance for each car. Defaults to 5. Must be a float between 1 and 100000.
--shared_auction_parameters			Boolean	All auctions will share parameters, instead of each auction having its own parameters.
--adaptive_auction_action_selection	e_greedy_exp_decay	[boltzmann, e_greedy_decay, e_greedy_exp_decay, ucb1, reverse_sigmoid_decay, random, zero]	str	Type of auction modifier action selection. Defaults to 'e_greedy_exp_decay'. Must be one of the listed options.
--bid_calculation_rule	linear	[linear, multiplicative, non-linear]	str	This is the way the bid is calculated. Must be one of 'linear', 'multiplicative', or 'non-linear'.
--adaptive_auction_update_rule	simple_bandit	[simple_bandit, svr]	str	This is the rule used to update the expected reward for each parameter combination. Defaults to simple_bandit. Must be one of 'simple_bandit' or 'svr'.
--auction_episode_length	10	[1-1000]	int	Length of an episode for the adaptive auction. Defaults to 10. Must be an integer between 1 and 1000.
--action_selection_hyperparameters			float	Hyperparameters to use for the action selection algorithm (consult documentation for details).
--adaptive_auction_discretization	25	[1-1000]	int	Number of discrete values to check for each parameter of the adaptive auction. Defaults to 25. Must be an integer between 1 and 100.
--only_winning_bid_moves			Boolean	If enabled, only the car with the winning bid will move.
--intersection_reward_type	mixed_metric_rank	[inact_rank, rank_dist_metric, mixed_metric_rank, mixed_rank_dist_metric]	str	Type of reward for the intersection. Must be one of the listed options.
--inact_rank_weight	0.5	0-1	float	Weight of the inactivity rank used for the intersection auction reward. Defaults to 0.5. Must be a float between 0 and 1.
--bid_rank_weight	0.5	0-1	float	Weight of the bid rank used for the intersection auction reward. Defaults to 0.5. Must be a float between 0 and 1.
--all_cars_bid			Boolean	If enabled, all cars of a queue can bid, not only the first one.
--shared_bid_generator			Boolean	If enabled, all bidders will share the same bid generator.
--bidders_proportion	[1, 0, 0, 0, 1]		int	Proportion for the types of bidders to use. Defaults to [1, 0, 0, 0, 1] (only homogeneous and RL).
--bidders_urgency_distribution	gaussian	[gaussian, beta]	str	Type of distribution from which the bidders' urgency is sampled. Defaults to 'gaussian'. Must be one of the listed options.
--results_folder	results/YYYY-MM-DD_HH-MM-SS		str	Path to the Results Folder. Defaults to 'results/' + current date & time.
--print_grid			Boolean	If present, the grid will be printed after each epoch.
--sweep_mode			Boolean	If present, no plots are generated to reduce execution time.
--low_dpi			Boolean	If enabled, all plots are created with low dpi to reduce execution time.

Cleaning Previous Results

To clean the results of previous simulations (including cache), simply run:

python aatc.py clean