Heterogeneous-Highway-Env 0.0.3

Codebase for for our version of highway-way modified for iPLAN paper

Heterogeneous Highway-env for iPLAN

Our forked version of highway-env used in our paper:

iPLAN: Intent-Aware Planning in Heterogeneous Traffic via Distributed Multi-Agent Reinforcement Learning

More details could be found in the main page of highway-env and iPLAN.

Installation

pip install HighwayEnv-iPLAN

Major Changes

• Add two behavior-driven vehicle models, DefensiveVehicle and AggressiveVehicle in vehicle/behavior.py.
• Add multi-agent support for Highway scenario given in envs/highway_env.py, modify the MultiAgentWrapper in vehicle/common/abstract.py.
• Add three heterogeneous traffic scenarios, HighwayEnvHetero, HighwayEnvHetero_H and HighwayEnvHetero_VH in envs/highway_env.py, with vehicle ID broadcasting and different behavior-driven vehicles.
• Add multi-agent support for visualization in Highway scenario that allows a camera following each agent and visualize their surroundings from their respective viewpoints.

Animation

The animation shows 5 such learning agents (Green) with their surroundings from their respective viewpoints. Behavior-driven vehicles in the environment include: Normal (Blue), Aggressive (Purple) and Defensive (Yellow). Vehicles terminate (Red) when colliding with other vehicles.

iPLAN in chaotic (hard) scenario of Heterogeneous Highway (Num of agents succeed: 5, Avg. survival time: 90, Avg. speed: 21.81).