Hi, I am trying to run this code with Mujoco Pro license in OSX but I am getting this ERROR: Could not open disk Code is import gym env = gym.make('Hopper-v1') env.reset() env.render() I have setup the mjpro131 properly, I had mjpro150 but the openAI code gave an error that 1.31 should be used so downloaded and placed the mjkey in bin and ran but failed, so followed the steps mentioned in Mujoco documentation and I am still getting the same error "ERROR: Could not open disk"
This happens on Macs with NVMe disks and MuJoCo versions prior to 1.50. It was fixed in 1.50, but there is no way to go back in time and fix it in older releases... unless I misunderstand your post... are you saying you got this error with 1.31 or 1.50?
1.50 works fine and its amazing, I was trying OpenAI Mujoco models which needs 1.31 version but as you said it can't be fixed, thank you for the help professor
I wonder if there is a way to "go back in time", i.e., release a mujoco 1.32 that fixes the NVMe disk support... then OpenAI could easily support it (users could even just rename, or link, mjpro132 to mjpro131). I can see you might want to pressure OpenAI to support 1.50... but we (NVMe disk owners) are suffering! And that is an actual bug in mujoco 1.31... Thanks for considering it. @Gautam @Emo Todorov
Hi Mike, That would be a reasonable request for a larger company, but Roboti is effectively a one-person company, and the only way I can manage everything (on top of running a research group at UW) is to have a strict no-regret policy: there is only one code base, only one version of each release, and time only moves forward. Ideally there would be a bigger base of paying customers which would make it possible to hire additional developers as well as support staff, but at least for now that is not the case. So I cannot open the door to customization requests for prior releases. Suggestions for improvement in future releases are always welcome of course. Given that there is a substantial community of students using the software for free (1846 as of now), one would hope that they would start contributing simulation models. The OpenAI environments are not difficult to adapt to MuJoCo 1.50, and it is likely that someone somewhere has already done it - so it is just a matter of posting it on the Forum under Resources. Yours, Emo
Thank you for the very quick reply. I can respect this policy, one has to set some limits to remain sane! Yes there are some folks attempting to do this at the mujoco-py github page, for example this thread: https://github.com/openai/mujoco-py/issues/80 Cheers, Mike