search

OSUrobotics / KinovaGrasping

This contains the simulation of a kinova robot and the code for collecting data and training both a grasp classifier and a RL agent
25 stars 6 forks source link

Actor is fine regardless of state_dim #49

Open jimzers opened 3 years ago

jimzers commented 3 years ago

https://github.com/OSUrobotics/KinovaGrasping/blob/5333116a3ddc9315e6a4d3e718e9d49f40a12e01/gym-kinova-gripper/DDPGfD.py#L94

Works with both observation shapes of 82 and 26.

The input into the actor: (1, 82) and (1, 26).

Expected behavior: should throw an error when the wrong observation shape on the last axis is fed to the agent.

This may highlight deeper issues in our networks?