Reward Penalties on Augmented States for Solving Richly Constrained RL Effectively

Authors: Hao Jiang, Tien Mai, Pradeep Varakantham, Huy Hoang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experimental Results We experimentally answer the following questions with regards to our approaches :
Researcher Affiliation Academia Singapore Management University haojiang.2021@phdcs.smu.edu.sg, atmai@smu.edu.sg, pradeepv@smu.edu.sg, mhhoang@smu.edu.sg
Pseudocode Yes The pseudo code for the Safe DQN algorithm is provided in the appendix. [...] The detailed pseudocode for Safe SAC is provided in the appendix.
Open Source Code No The paper does not contain any statement or link indicating that the source code for their methodology is openly available.
Open Datasets Yes For a discrete state and discrete action environment, we consider the stochastic 2D grid world problem introduced in previous CMDP works (Leike et al. 2017; Chow et al. 2018; Satija, Amortila, and Pineau 2020; Jain, Khetarpal, and Precup 2021). [...] Next, we consider the highway environment [...] (Leurent 2018). [...] We then compare Safe SAC with recent safe methods for continuous action spaces on the two environments Safety Point Goal1-v0, Safety Car Goal1-v0 from Safety Gymnasium (Ji et al. 2023).
Dataset Splits Yes The performance values (expected cost and expected reward) along with the standard deviation in each experiment are averaged over 5 runs.
Hardware Specification No The paper does not explicitly describe the hardware (e.g., specific GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper does not provide specific version numbers for software dependencies (e.g., Python, PyTorch, or other libraries).
Experiment Setup Yes We set the expected cost threshold, cmax = 2, meaning agent could pass at most one pit. [...] We set the cmax = 8. [...] We set cmax = 15. [...] we conduct experiments on Grid World using Safe SAC, with λ1 = 1, λ2 = 5λ1, λ3 = 10λ1, a small λ4 = 0.001 and λ5 = 0