Open benelot opened 6 years ago
Hi, first of all, thank you so much for this massive work to make everything open-source. This comes in hand especially in this period where everyone is forced to work from home and it becomes very difficult to train remotely your virtual robot with the MuJoCo Licence.
I think that a good integration would be to make a pybullet version of the robosuite's robots. On github I found the prototyping URDFs for pybullet for the same robots but not their environments ready to be used with the openai algorithms.
Do you think this is a possible step for the future?
Hi!
I heard of the problems to train mujoco envs with mujoco from remote. The env files of robosuite seem to be there, but it would take some time to make them load properly as our pybullet interface is not yet very mature to just load them as-is. If you are willing to contribute to that, then I can give you guidance. Especially if you are experienced with the robosuite, that could really help.
On Tue, Jun 2, 2020 at 1:17 AM mart notifications@github.com wrote:
Hi, first of all, thank you so much for this massive work to make everything open-source. This comes in hand especially in this period where everyone is forced to work from home and it becomes very difficult to train remotely your virtual robot with the MuJoCo Licence.
I think that a good integration would be to make a pybullet version of the robosuite's robots https://github.com/StanfordVL/robosuite. On github I found their prototyping URDFs for pybullet https://github.com/erwincoumans/pybullet_robots but not their environments ready to be used with the openai algorithms.
Do you think this is a possible step for the future?
— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/benelot/pybullet-gym/issues/1#issuecomment-637175329, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAXXXK3DJVT4IQXS46EZ5TDRUQZI7ANCNFSM4ELUSNYQ .
Hey Benjamin Thank you for taking the time to respond to me. Yes, I would love to contribute if that's possible.
First off, it needs to be compatible with agents trained on MuJoCo trained envs. Any contribution welcome of such agents! Just post here if you have a set of well trained MuJoCo task agents.
If somebody even has a trace of observations together with the agent, that could be super helpful to find the differences.