Norm Conflict Resolution in Stochastic Domains

Authors: Daniel Kasenberg, Matthias Scheutz

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide a proof-of-concept implementation in a simulated vacuum cleaning domain. We implemented the proposed approach in BURLAP (Mac Glashan 2016), a Java library for MDP planning and reinforcement learning (RL). We tested the algorithms in four different scenarios in the vacuum cleaning example which use one, two, three, or all four norms (N1-N4).
Researcher Affiliation Academia Daniel Kasenberg, Matthias Scheutz Human-Robot Interaction Laboratory Tufts University, Medford, MA, USA
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper mentions using BURLAP, a Java library, and Rabinizer 3 for conversion from LTL formulas to DRAs, both with citations to papers describing them. However, it does not state that the specific code for the methodology or its implementation in the simulated vacuum cleaning domain is open-source or publicly available.
Open Datasets No The paper uses a custom simulated vacuum cleaning domain for its experiments. It does not refer to any publicly available dataset.
Dataset Splits No The paper describes a simulated environment and scenarios with varying parameters but does not specify training, validation, and test dataset splits in the conventional sense, as it's a simulation rather than a typical machine learning dataset.
Hardware Specification No The paper does not explicitly describe the specific hardware used to run its experiments.
Software Dependencies No The paper mentions "BURLAP (Mac Glashan 2016)" and "Rabinizer 3 (Esparza and Ket ınsk y 2014)". While these are software names, the provided text does not specify version numbers for BURLAP or Rabinizer 3 (e.g., BURLAP vX.Y), only the year of the associated publication.
Experiment Setup Yes In each scenario, the robot has 10 units of health (the robot s battery capacity varies between scenarios). The human begins in Room 2. The probability of the human transitioning between rooms at each time step is 0.125. The human creates messes in their current room with probability 0.2 in each time step (except in Scenario 4, in which the human does not create new messes). All messes created by the human are harmless and do not damage the robot, and initially have 2 units of dirtiness. In each case, we set the discount factor γ = 0.99.