rr-learning / CausalWorld

CausalWorld: A Robotic Manipulation Benchmark for Causal Structure and Transfer Learning
https://sites.google.com/view/causal-world/home
MIT License
212 stars 26 forks source link

Torque-mode wrong achieved_goal #80

Closed martius-lab closed 4 years ago

martius-lab commented 4 years ago

The 'get_achieved_goal' function, in e.g. reaching, returns the end_effector_positions from the '_latest_full_state' dict. This is not updated in 'joint_torque' mode, giving constant values for the achieved goal.

I suggest to split the purpose of the last_full_state. There should be a structure to store the endeffector position: current_state that is updated just once every step (I think in do_simulation() after the physics update)

then the storage of the last_full_state and its update is a different issue that I report in a moment.