search

Jasonxu1225 / Uncertainty-aware-Inverse-Constrained-Reinforcement-Learning

[ICLR 2024] ''Uncertainty-aware Constraint Inference in Inverse Constrained Reinforcement Learning'' Official Code
MIT License
8 stars 1 forks source link

How could we inject noise in highD environments? #3

Closed Usaywook closed 2 weeks ago

Usaywook commented 2 weeks ago

I found that there are no differences between config/HighD_velocity_constraint/train_Binary_highD_velocity_constraint.yaml and config/HighD_velocity_constraint/train_Binary_highD_velocity_constraint-1e-1.yaml

Other files to implement algorithms in highD environments have the same issues.

How do we inject noise in highD environments to reproduce the result like Figure D.10 in the paper?

Jasonxu1225 commented 2 weeks ago

As the paper said, in HighD env we "add Gaussian noise to the agent’s action $a_t$ to introduce the control-level stochasticity". So you should find it in the commonroad_environment/commonroad_rl/gym_commonroad/action/action.py. Specifically, in line 260. You can modify it as you need. I kindly recommend reviewing the code first, rather than asking questions without exploring it.

Usaywook commented 2 weeks ago

Thank you for guiding me through regulating the transition noise in a high-D environment. Your answer solved my concern.

Jasonxu1225 commented 2 weeks ago

You're very welcome. I'm glad the explanation helped resolve your concern. I'm always happy to help.