Performance Analysis of the Jason Reasoning Cycle

Abstract

Jason is a popular interpreter of AgentSpeak(L) that provides an environment to develop and run autonomous agents. We are interested in determining if Jason can be used to develop autonomous robots that can be run on modern hardware. To this end, we determined the time complexity of each of the ten steps of the Jason reasoning cycle and used that complexity to help us identify what parameters could be changed to have the biggest effect on the execution time of the reasoning cycle.We also generate a model to predict future performance. We validate this model using three different case studies, comparing the predicted execution times to those we measure. We find that our developed model consistently predicted an execution time one order of magnitude higher than the measured execution time for each case study.We further find that, using modern hardware, Jason provides a suitable environment for autonomous robots. i I would like to first and foremost sincerely thank my thesis supervisor, Professor Babak Esfandiari. He has given me the latitude to pursue this research in the direction I desired, while providing extremely insightful suggestions and questions to ensure I didn't get lost along the way. Special thanks to Patrick Gavigan, who was instrumental in collecting data from hardware that I didn't have access to. I also wish to express my deep gratitude to my wife, Annie St-Amour, who has supported me every step of the way. Annie, your belief in me kept me going when I didn't think I could. Finally, thank you to my sponsors. We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC) and Four DRobotics. Nous remercions le Conseil de recherches en sciences naturelles et en génie du Canada (CRSNG) ainsi que Four DRobotics de leur soutien.

Qualitative Uncertainty Reasoning in AgentSpeak

“…A naive approach involves propositionalizing all logical consequences of rules existing in the belief base, ensuring that the reasoner has access to every propositional form of all belief rules during model creation. However, the logical consequences function is computationally expensive, requiring O(|B| |R| ) time and space in the worst-case scenario for B total belief queries used within R recursive rule consequences [51]. Obtaining the propositionalization of all rule consequences would significantly and unnecessarily impact performance.…”

“…The worst-case computational complexity for the standard logical consequences function must account for the potential recursive evaluation of belief rules R and rule bodies containing beliefs B; thus the worst-case time and space complexity is O(|B| |R| ) [51]. In practice, this depends heavily on how many literals require nested evaluation of rules.…”