DeepTraffic: Crowdsourced Hyperparameter Tuning of Deep Reinforcement Learning Systems for Multi-Agent Dense Traffic Navigation

TL;DR

DeepTraffic is an accessible traffic simulation platform designed to facilitate large-scale, crowdsourced hyperparameter tuning of deep reinforcement learning policies for multi-agent traffic navigation, fostering education and research.

Contribution

This paper introduces DeepTraffic, a novel traffic simulation environment that enables crowdsourced hyperparameter optimization for deep RL in multi-agent traffic scenarios.

Findings

Successful large-scale hyperparameter tuning through crowdsourcing

Enhanced understanding of hyperparameter impacts on deep RL performance

Engagement of thousands of participants in RL research

Abstract

We present a traffic simulation named DeepTraffic where the planning systems for a subset of the vehicles are handled by a neural network as part of a model-free, off-policy reinforcement learning process. The primary goal of DeepTraffic is to make the hands-on study of deep reinforcement learning accessible to thousands of students, educators, and researchers in order to inspire and fuel the exploration and evaluation of deep Q-learning network variants and hyperparameter configurations through large-scale, open competition. This paper investigates the crowd-sourced hyperparameter tuning of the policy network that resulted from the first iteration of the DeepTraffic competition where thousands of participants actively searched through the hyperparameter space.