Search papers, labs, and topics across Lattice.
This paper introduces a robot path planning strategy for dynamic environmental monitoring using spatiotemporal Gaussian Process Regression (GPR) to model environmental conditions and their associated uncertainties over time. The proposed Predictive Uncertainty Minimization (PUM) method guides the robot's trajectory to minimize future uncertainties predicted by the GPR model. Simulation results using real-world environmental data demonstrate that PUM achieves lower uncertainty levels compared to predefined path and highest-uncertainty-point-seeking methods, reducing overall uncertainty by up to 9.25%.
Robots can monitor dynamic environments more effectively by planning paths that proactively minimize future uncertainty, outperforming traditional approaches by up to 9.25%.
In environmental monitoring, traditional methods primarily focus on collecting as much spatial information as possible to construct a more accurate static model of the environment. However, these methods fail to account for the fact that environmental factors, such as temperature, can vary over time during monitoring, making it difficult for robots efficiently capture temporal dynamics of each region across the entire spatial area. This paper addresses these challenges by applying a spatiotemporal Gaussian Process Regression (GPR) model, which estimates not only the environmental conditions of unobserved regions but also their associated uncertainties based on observed data, considering time variations. Building upon this model, we propose a method called Predictive Uncertainty Minimization (PUM), which controls the robot's trajectory to minimize future uncertainties. Through simulation using the data set obtained in the real-world environment, we show that PUM consistently maintains lower uncertainty levels, reducing overall uncertainty by 7.91% and 9.25% compared to the method following a predefined path and the method continuously moving to the point with the highest uncertainty, respectively.
