Location
FCRAR 2023 Tampa
Event Website
https://digitalcommons.usf.edu/fcrar/
Start Date
19-5-2023 10:40 AM
End Date
19-5-2023 11:00 AM
Document Type
Full Paper
Keywords
POMDPs, Planning
Description
We compare the relevance of POMDP graphical models to addressing a given common task, the autonomous navigation, in mobile robotics, for robots in three different environments: legged walking, wheeled, and underwater robots. Principled approaches to the implementation of Intelligent robots in realistic environments with uncertainty exploit probabilistic models. The use of graphical models increases tractability and verifiability of probabilistic reasoning. Our robots face similar localization problems of passing by three doors on a corridor in a building, respectively, for the underwater case, of passing by three barrels in a pool. All the three robots are equipped solely with noisy sonars and odometers, a natural restriction for the underwater case, and a significant explanation of the encountered level of uncertainty. POMDP models are proposed. The experiments performed show that when the three robots handle their respective problem, the legged robot has the highest locomotion challenges, while the underwater setting faced the highest level of noise in sensors.
DOI
https://doi.org/10.5038/TVGT2918
Comparing probabilistic planning success on three types of robots: wheeled, legged, and underwater
FCRAR 2023 Tampa
We compare the relevance of POMDP graphical models to addressing a given common task, the autonomous navigation, in mobile robotics, for robots in three different environments: legged walking, wheeled, and underwater robots. Principled approaches to the implementation of Intelligent robots in realistic environments with uncertainty exploit probabilistic models. The use of graphical models increases tractability and verifiability of probabilistic reasoning. Our robots face similar localization problems of passing by three doors on a corridor in a building, respectively, for the underwater case, of passing by three barrels in a pool. All the three robots are equipped solely with noisy sonars and odometers, a natural restriction for the underwater case, and a significant explanation of the encountered level of uncertainty. POMDP models are proposed. The experiments performed show that when the three robots handle their respective problem, the legged robot has the highest locomotion challenges, while the underwater setting faced the highest level of noise in sensors.
https://digitalcommons.usf.edu/fcrar/2023/May19/3