Intention Progression with Temporally Extended Goals

Authors: Yuan Yao, Natasha Alechina, Brian Logan

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate our approach using a version of the Mars Rover domain [Duff et al., 2006], in which the tasks assigned to the Rover are specified using temporally extended goals.
Researcher Affiliation Academia Yuan Yao1 , Natasha Alechina2,3 , Brian Logan4,3 1University of Nottingham Ningbo China 2 Open University Netherlands 3Utrecht University 4University of Aberdeen yuan.yao@nottingham.edu.cn, {n.a.alechina, b.s.logan}@uu.nl
Pseudocode Yes Algorithm 1 Return the step to be executed at this cycle
Open Source Code Yes The software used in the experiments is freely available at https: //github.com/yvy714/TEG-MCTS.
Open Datasets No The Mars Rover domain is a two-dimensional grid environment where the agent (the rover) can move around and conduct soil experiments at different locations (see Figure 5)... The locations for conducting soil experiments and the holes are randomly generated. However, the base and the experiment locations are not located in a hole.
Dataset Splits No The paper describes experimental runs in a simulated environment ('average performance over 50 runs') but does not specify train/validation/test dataset splits as it's not a traditional dataset-based machine learning task.
Hardware Specification Yes our prototype implementation requires about 50 milliseconds to return the best next plan for 15 concurrent intentions on a 1.4 GHz Quad-Core Intel Core i5
Software Dependencies No The paper mentions 'The software used in the experiments is freely available at https: //github.com/yvy714/TEG-MCTS' but does not specify any particular software components with version numbers.
Experiment Setup Yes In each deliberation cycle, the scheduler performs 100 iterations (α = 100) and 10 simulations per iteration (β = 10).