search

Meowuu7 / QuasiSim

[ECCV'24] Parameterized Quasi-Physical Simulators for Dexterous Manipulations Transfer
https://meowuu7.github.io/QuasiSim/
MIT License
56 stars 3 forks source link

The Generalization ability to different video sequences #4

Closed zhiyuanYU134 closed 1 week ago

zhiyuanYU134 commented 1 week ago

What a great job！But I want to know whether the model weights can generalize for different video sequences. And can 'Parameterized Quasi-Physical Simulator' simulate multiple environments at the same time like issac gym?

Meowuu7 commented 1 week ago

Hi @zhiyuanYU134 ,

Thank you for your interest. Since QuasiSim infers actions through per-trajectory optimization rather than training neural networks, the pre-trained weights are trajectory-specific and cannot generalize to different video sequences. Currently, the 'Parameterized Quasi-Physical Simulator' does not support multiple environments or parallel simulations. What sets it apart from mainstream simulators is its tunability, allowing for varying levels of optimizability and fidelity.

I hope this clarifies your questions. Thank you!

zhiyuanYU134 commented 1 week ago

ｔｈａｎｋｓ　ｆｏｒ　ｙｏｕｒ　ｉｍｍｅｄｉａｔｅ　ｒｅｐｌｙ！ｙｏｕｒ　ａｎｓｗｅｒ　ｉｓ　ｈｅｌｐｆｕｌ．