Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

apparent environment seed issues with ant environment #72

Open
balisujohn opened this issue Sep 6, 2021 · 0 comments
Open

apparent environment seed issues with ant environment #72

balisujohn opened this issue Sep 6, 2021 · 0 comments

Comments

@balisujohn
Copy link

balisujohn commented Sep 6, 2021

The following code checks to see if when supplied with randomly sampled actions sampled from two action spaces supplied with the same seed, two instances of the pybullet ant environment will create the same observations. This code seems to fail intermittently in python3.5 and consistently in python3.6. For the life of me, I can't figure out what is causing the drift between environment instances.

System Specs:

Ubuntu 18.04
python 3.6 (also verified with python 3.5)



import gym
import pybulletgym
import numpy as np



if __name__ == "__main__":
    
    env1 = gym.make("AntPyBulletEnv-v0")
    env1.seed(0)
    env1.action_space.seed(0)

 

    env2 = gym.make("AntPyBulletEnv-v0")
    env2.seed(0)
    env2.action_space.seed(0)

    obs1 = env1.reset()
    obs2 = env2.reset()
    for i in range(100):
        if not np.array_equal(obs1 ,obs2 ):
            for e1,e2 in zip(obs1,obs2):
                if e1 != e2:
                    print(e1,e2)
            exit("failed on obs")
        action1 = env1.action_space.sample()
        action2 = env2.action_space.sample()
        if not np.array_equal(action1, action2):
            print(action1, action2)
            for a1,a2 in zip(action1,action2):
                if a1 != a2:
                    print(e1,e2)
            exit("failed on action")
        print("env 1")
        obs1, reward, done1, info = env1.step(action1)
        print(action1, obs1)
        print("env 2")
        obs2, reward, done2, info = env2.step(action2)
        print(action2, obs2)
        if done1:
            assert(done2)
            if not np.array_equal(obs1 ,obs2 ):
                for e1,e2 in zip(obs1,obs2):
                    if e1 != e2:
                        print(e1,e2)
                exit("failed on obs")
            obs1 = env1.reset()
            obs2 = env2.reset()
       
## Output
...
env 1
[-0.19167283 -0.24867578  0.57644254 -0.6455737   0.5354068   0.95332575
 -0.48177752  0.2555853 ] [-0.22058617 -0.07561598  0.997137    0.10426999 -0.00999914  0.11529232
  0.162765   -0.11414797 -1.0032024   0.16360427  0.50044924  0.0312518
 -0.58839995  0.32609588 -0.03842217 -0.13290787  0.2856137   0.26151842
  1.1269124   0.05937529 -0.35873523  0.02834896 -0.61619514  0.8499997
  1.          0.          0.          0.        ]
env 2
[-0.19167283 -0.24867578  0.57644254 -0.6455737   0.5354068   0.95332575
 -0.48177752  0.2555853 ] [-0.22060278 -0.07551313  0.9971448   0.10177492 -0.01480584  0.11399316
  0.1625488  -0.11415483 -1.0028749   0.17036478  0.49990714  0.03061081
 -0.58719516  0.33724806 -0.03812427 -0.13370141  0.28572914  0.2649531
  1.1260623   0.05638258 -0.3578396   0.03535499 -0.6153823   0.8513867
  1.          0.          0.          0.        ]
-0.22058617 -0.22060278
-0.07561598 -0.07551313
0.997137 0.9971448
0.10426999 0.101774916
-0.009999137 -0.014805844
0.11529232 0.11399316
0.162765 0.1625488
-0.11414797 -0.11415483
-1.0032024 -1.0028749
0.16360427 0.17036478
0.50044924 0.49990714
0.031251803 0.030610807
-0.58839995 -0.58719516
0.32609588 0.33724806
-0.038422175 -0.03812427
-0.13290787 -0.13370141
0.2856137 0.28572914
0.26151842 0.2649531
1.1269124 1.1260623
0.059375294 0.056382578
-0.35873523 -0.3578396
0.028348956 0.035354994
-0.61619514 -0.6153823
0.8499997 0.8513867
failed on obs
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant