Reinforcement Learning by Phil Winder Ph. D

Author:Phil Winder Ph. D. [D., Phil Winder Ph.]
Language: eng
Format: azw3
Publisher: O'Reilly Media
Published: 2020-11-05T16:00:00+00:00

Description of the Problem

The state is represented by an observation through a video camera. The video has a resolution of 160×120 with 8-bit RGB color channels. The action space is deceptively simple, allowing a throttle amount and a steering angle as input. The environment rewards the agent for speed and staying in the center of a lane/track on every time step. If the car is too far outside its lane it receives a negative reward. The goal is to reach the end of the course in the fastest time possible.

There are several tracks. The simplest is a desert road with clearly defined and consistent lanes. The hardest is an urban race simulation, where the agent has to navigate obstacles and a track without clear boundaries.

The Unity simulation is delivered as a binary. The Gym environment connects to the simulation over a websocket and supports multiple connections to allow for RL-driven competitive races.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.