search

czp16 / FCSRL

Feasibility Consistent Representation Learning for Safe Reinforcement Learning (ICML 2024). Current SOTA model-free safe RL algorithm on safety-gym
https://sites.google.com/view/fcsrl
Apache License 2.0
5 stars 0 forks source link

a small question #1

Closed Tietang999 closed 2 months ago

Tietang999 commented 2 months ago

Your paper is excellent,i have a small problem with it,why use TD-lambda to compute return and costs.thanking for your answer.

czp16 commented 2 months ago

Thanks for your question. TD-lambda can reduce the variance in value function estimation (both for reward and cost). For fair comparison, we also use TD-lambda for all representation learning baselines based on TD3-Lag.

Tietang999 commented 2 months ago

Thank you for your reply. I wish you success in your research and a happy life!