Great job! I am a noob. Do you get any idea to train fetchpickandplace-v1 using the same algorithm?
Closed this issue · 8 comments
Great job! I am a noob. Do you get any idea to train fetchpickandplace-v1 using the same algorithm? some thing i really need to pay attention , and the difference between the two tasks. I tried to train it on the basis of your code. but didn't make it. Thanks in advance!
Hi. I've never tried fetchpickandplace-v1. Pick and place seems to have more observation spaces and action spaces. You might have to tweak the code. If the code is working but it doesn't converge, I would say that you might want to try different neural net architectures (e.g. more layers, convolutional, etc.). Let me know if you have further questions. Good luck!
I tried several time, but still couldn't make it . Would you like to try fetchpickandplace-v1 with the same algorithm? Considering you have done fetchreach so perfect, it cannot be difficult for you.
Hi Chao-Hou. I don't have a plan to do fetchpickandplace-v1 at this point. But I'll be glad to help anyone. Thanks.
Hi pipatth, Thanks for your awesome presentation, I would be glad if you can help me because am finding it difficult to understand some basic implementation process.
Thanks in anticipation my email(gbenga28@gmail.com)
Hi @pipatth , did you manage to implement for PickAndPlace ?, if so can you share the code ? Thank you! mail: bandi.chai414@gmail.com
@chaiban I haven't try PickAndPlace. It shouldn't be much different than Fetch. You need to give the learner some rewards along the way (hindsight experience replay). Otherwise, the learning is so slow.
@chaiban I haven't tried customer environment on Mujoco before. Have you tried looking at https://github.com/openai/gym/tree/master/gym/envs/mujoco ?