search

ugo-nama-kun / gym_torcs

MIT License
408 stars 163 forks source link

Are low-dimensional state observations different between with vision and without vision? #40

Open JunhongXu opened 7 years ago

JunhongXu commented 7 years ago

I want to train an agent using DDPG with low-dim input and use it as a teacher to train a imitation learning agent. However, after training DDPG, when I set vision=True, I found DDPG can not perform well. But I set vision=False back, DDPG can drive without any issue.

My question is: are the scales or measurements of the low-dim state different between vision=True and vision=False.

BCWang93 commented 6 years ago

你好，我想问一下。你在使用图像作为输入的时候，程序运行是正常的吗？我这里总是出现错误。。谢谢！

colinzhaoxp commented 2 years ago

你好，我想问一下。你在使用图像作为输入的时候，程序运行是正常的吗？我这里总是出现错误。。谢谢！

你好，我现在把程序设置为64x64，但是获取的64x64图像都是黑的，请问问题您解决了吗？