DDPG never converge

[iswaverly]

I trained DDPG with command refered in README, which is "python -m baselines.run --alg=ddpg --env=HalfCheetah-v2 --num_timesteps=1e6".
After 1000000 steps, the reward is still negative.
I tried other games, such as "popper", none of this can I get the correct result.

I use the master branch with latest code, and my tensorflow-gpu version is 1.8.0.

Have anyone train DDPG success?

[iswaverly]

fix it

[Yeosangho]

Can you share your solution with us? I have same prorblem like you. TT

[sayomakinwa]

I have a similar situation with DDPG on Humanoid-v2 environment; it doesn't converge. Suggestions will be highly appreciated

[DennisWangCW]

I meet with the same situation. Hope for solution.

[DennisWangCW]

fix it

hi, could you share with us how you fix it?

[fangchuan]

HI, I have same problems when I implement my agent based on baselines.ddpg, counld you tell us how you fix the unconverged problem? I really appreciate that.