search

watakandai / hiro_pytorch

Implementation of HIRO (Data-Efficient Hierarchical Reinforcement Learning)
90 stars 20 forks source link

When I train with the original parameters, the success rate is 0, what should I do? #8

Open chenci107 opened 3 years ago

chenci107 commented 3 years ago

I used the original parameters in the code for training. I trained 24,000 steps, and the success rate is always 0 during the test. What should I do to improve the success rate?

Roythuly commented 2 years ago

Which environment you used? I try the AntMaze, it can work well by nearly 3M steps. But I also cannot work on AntPush and AntFull.