Vehicle Environment Static Obstacles with Unity ML-Agent
This simulator is simple vehicle environment. In this environment, Agent should evade obstacle and get stars while keeps center of the lane.
Sample video of this game is as follows.
Above demo, referenced papers to implement algorithm are as follows.
Noisy Networks for Exploration (arXiv Link)
Deep Reinforcement Learning with Double Q-learning (arXiv Link)
The download links of this environments are as follows.
This environment is made by purchased assets, so raw unity code cannot be provided.
The links of the built program are as follows.
Car Game Static Obs Windows Link
Car Game Static Obs Mac Link
Car Game Static Obs Linux Link
Link of the asset are as follows.
Simple Racer Asset Link
Simple Town Asset Link
Objects of this environment are as follows.
Agent: red vehicle in the camera center
Obstacles: red and white tire barriers which randomly generates at the lane center in every episode
stars: stars which randomly generates at the lane center in every episode
Actions of the environments are as follows.
Move Left to a certain distance
Move Right to a certain distance
Stay current state
Reward of the environments are as follows.
As above image, if agent hits star, it gets +5 reward. However, agent hits tire barriers, it gets -2.5 reward. In the case of the lane, agent gets +0.1 at the center of every lane. If agent gets far from lane center, reward linearly decrease to 0.
According to the reward, the goal of the agent is keep center of the lane and gets stars while evading tire barriers.
Terminal Conditions of the environments are as follows.