Have trouble running the pretrained policy

[RyanJiao]

I finished setting up the envirment and run the code _python3 train.py wandb_activate=false numenvs=1 headless=false test=true task=LeapHandRot checkpoint=runs/pretrained/nn/LeapHand.pth and then error ocurred : FileNotFoundError: [Errno 2] No such file or directory: 'cache/leap_hand_in_palm_cube_grasp_50k_s095.npy' How can I fix it?

[RyanJiao]

[leap-hand]

Hi @RyanJiao

Thanks for pointing this out! We just pushed the grasp cache, please check the most recent commit 0e9f0a3. This should hopefully fix the issue.

[RyanJiao]

Thank you so much for sharing such a good project! This should help fix my problem. But I’d like to ask if you have ever encountered such faults during training in Isaac gym:

pxgcudadevicememoryallocator fail to allocate memory 67108864 bytes!! result = 2

It will be really nice if you can give me any hints on it.  Many thanks!

------------------ Original ------------------ From: leap-hand @.> Date: Tue,Sep 19,2023 0:28 AM To: leap-hand/LEAP_Hand_Sim @.> Cc: Ryan Jiao @.>, Mention @.> Subject: Re: [leap-hand/LEAP_Hand_Sim] Have trouble running the pretrainedpolicy (Issue #1)

Hi @RyanJiao

Thanks for pointing this out! We just pushed the grasp cache. This should hopefully fix the issue.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

[anag004]

What GPU are you using to run the code? Seems like the vRAM might not be enough. You can try running on CPU by specifying pipeline=cpu.

python3 deploy.py wandb_activate=false num_envs=1 headless=false test=true task=LeapHandRot checkpoint=runs/pretrained/nn/LeapHand.pth pipeline=cpu

Another thing to try is to turn off VHACD here (just change True to False) https://github.com/leap-hand/LEAP_Hand_Sim/blob/0e9f0a3de7f38281f71ba88e5dade4d6a6586dac/leapsim/tasks/leap_hand_rot.py#L981

[RyanJiao]

Hi! while excuting this: I encountered a new fault. Do you any clue to solve this? Thanks again for your help! for cube_scale in 0.9 0.95 1.0 1.05 1.1 do bash scripts/gen_grasp.sh $cube_scale custom_grasp_cache num_envs=1024 done

焦睿轩 @.***

 

------------------ 原始邮件 ------------------ 发件人: "leap-hand/LEAP_Hand_Sim" @.>; 发送时间: 2023年9月19日(星期二) 凌晨3:24 @.>; @.**@.>; 主题: Re: [leap-hand/LEAP_Hand_Sim] Have trouble running the pretrained policy (Issue #1)

What GPU are you using to run the code? Seems like the vRAM might not be enough. You can try running on CPU by specifying pipeline=cpu.

python3 deploy.py wandb_activate=false num_envs=1 headless=false test=true task=LeapHandRot checkpoint=runs/pretrained/nn/LeapHand.pth pipeline=cpu

Another thing could be to turn off VHACD here (just change True to False) https://github.com/leap-hand/LEAP_Hand_Sim/blob/0e9f0a3de7f38281f71ba88e5dade4d6a6586dac/leapsim/tasks/leap_hand_rot.py#L981

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

[RyanJiao]

Hi! I only have a RTX3060 with 6GB memory, I run the pretrained policy by reducing the minibatch_size to 1 with num_envs=1, I wander if I can sucessfully train any policy on my RTX3060, If it can't make it, I will change my graphcard.  Many thanks!

焦睿轩 @.***

 

------------------ 原始邮件 ------------------ 发件人: "leap-hand/LEAP_Hand_Sim" @.>; 发送时间: 2023年9月19日(星期二) 凌晨2:25 @.>; @.**@.>; 主题: Re: [leap-hand/LEAP_Hand_Sim] Have trouble running the pretrained policy (Issue #1)

Hi @RyanJiao

I think you have an out of memory error, we used a RTX 3090 with 24GB vRAM. I would suggest you try decreasing the number of parallel environments in the yaml file.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

[anag004]

Hi! while excuting this: I encountered a new fault. Do you any clue to solve this? Thanks again for your help! for cube_scale in 0.9 0.95 1.0 1.05 1.1 do bash scripts/gen_grasp.sh $cube_scale custom_grasp_cache num_envs=1024 done 焦睿轩 @.   … ------------------ 原始邮件 ------------------ 发件人: "leap-hand/LEAP_Hand_Sim" @.>; 发送时间: 2023年9月19日(星期二) 凌晨3:24 @.>; @*.**@*.>; 主题: Re: [leap-hand/LEAP_Hand_Sim] Have trouble running the pretrained policy (Issue #1) What GPU are you using to run the code? Seems like the vRAM might not be enough. You can try running on CPU by specifying pipeline=cpu. python3 deploy.py wandb_activate=false num_envs=1 headless=false test=true task=LeapHandRot checkpoint=runs/pretrained/nn/LeapHand.pth pipeline=cpu Another thing could be to turn off VHACD here (just change True to False) https://github.com/leap-hand/LEAP_Hand_Sim/blob/0e9f0a3de7f38281f71ba88e5dade4d6a6586dac/leapsim/tasks/leap_hand_rot.py#L981 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.>

Hi, could you please comment directly on GitHub instead of through email? I think the image you sent was not included.

[RyanJiao]

OK！ This problem had been solved, Thank you!

[cqu-long]

好的！这个问题已经解决了，谢谢！

I am also trying to solve this problem. My launch file works well, but my deploy file does not. Can we talk about it together? Thank you