search

EvolutionGym / evogym

A large-scale benchmark for co-optimizing the design and control of soft robots, as seen in NeurIPS 2021.
https://evolutiongym.github.io/
MIT License
193 stars 33 forks source link

about observation and action vectors #19

Closed yyang8891 closed 2 months ago

yyang8891 commented 2 years ago

Hi all, I find the env simply outputs a vec of state observation and accepts action vec of actuators (mostly a subset of all voxels), while both of them (corner points and all voxels) are arrange in matrix format by default. So may I ask if anyone could please make it clear how are the two matrices converted into vectors? How do the 2-D coordinates correspond to vector indexes? Are they simply aligned from the top-left corner to down-right element like most of the cases in image processing treatments? I'll be very gratefull if anyone give me a hint, thanks in advance!

Yuxing-Wang-THU commented 1 year ago

Hi, you can refer to this page for more details about the observarion (https://github.com/Yuxing-Wang-THU/ModularEvoGym)

jagdeepsb commented 2 months ago

Hi @yyang8891 EvoGym does not make any guarantees about the order of corner points or order of actuators in the vectors except that the order doesn't change across environments and across time. The implementation I've used likely returns both in top-down, left-to-right order excluding repeats.