Collaborative Decision Making Using Action Suggestions

Authors: Dylan Asmar, Mykel J Kochenderfer

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We propose two methods that use suggested actions and demonstrate the approach through simulated experiments. The proposed methodology results in increased performance while also being robust to suboptimal suggestions.
Researcher Affiliation Academia Dylan M. Asmar Stanford Intelligent Systems Laboratory Stanford University Stanford, CA 94305 asmar@stanford.edu Mykel J. Kochenderfer Stanford Intelligent Systems Laboratory Stanford University Stanford, CA 94305 mykel@stanford.edu
Pseudocode No No pseudocode or algorithm blocks are present in the paper.
Open Source Code Yes Code is available at https://github.com/sisl/action_suggestions.
Open Datasets Yes The proposed methodology to incorporate action suggestions as observations was evaluated on two classic POMDP problems, Tag [30] and Rock Sample [38].
Dataset Splits No No explicit train/validation/test dataset splits are mentioned, as the experiments are conducted in simulated POMDP environments rather than on fixed datasets that are typically split for supervised learning.
Hardware Specification No Our approach does not require a lot of computation. Computation time and amount was not a concern and not provided.
Software Dependencies No The simulation environment was built using the POMDPs.jl framework [39].
Experiment Setup Yes Experiment details are documented in section 4. The hyperparameter is kept constant for each simulation and the value used is shown with the presented results.