Robotic Active Information Gathering for Spatial Field Reconstruction with Rapidly-Exploring Random Trees and Online...

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Networks

Background:

• Information gathering (IG) algorithms are crucial for accurate physical process reconstruction.
• Gaussian processes (GPs) are commonly used for IG but often require state-space discretization, limiting their applicability in complex robotic systems.
• Existing methods struggle with online IG tasks due to assumptions about GP parameters.

Purpose of the Study:

• To develop a novel, online IG algorithm for mobile sensors that overcomes the limitations of discretization and prior GP parameter knowledge.
• To enable efficient exploration and mapping of physical processes in dynamic environments.

Main Methods:

• A hybrid approach combining Rapidly-Exploring Random Tree (RRT) search for unvisited location identification and GP parameter learning.
• An RRT*-based informative path planning strategy to maximize gathered information while minimizing travel cost.
• Online adaptation and learning of GP parameters without prior knowledge.

Main Results:

• The proposed algorithm successfully performs online information gathering without state-space discretization.
• Demonstrated superior performance compared to state-of-the-art methods in both simulations and a real-world magnetic field mapping experiment.
• Efficiently navigated an indoor environment with obstacles.

Conclusions:

• The novel RRT and RRT*-based approach provides an efficient and adaptable solution for online information gathering with mobile sensors.
• This method significantly advances the capabilities of robotic systems in complex environments for tasks like mapping and exploration.
• Eliminates the computational burden of discretization and the need for pre-existing GP parameter knowledge.