Overview


Autonomous Mobile Robotics at URCS


---

The image on the right describes the overall picture of the proposed research. At the lowest level, we have sensors (and sensor models) which are used by the robot to acquire perceptions from the environment. This is described in Task 1.

The sensor manager then looks at the various sensor perceptions and guides the sensor fusion process. Sensor fusion is performed using Bayesian networks. Another important component of the sensor fusion task is the temporal belief manager. This module provides temporal dynamics and continuity to the sensor fusion module by using old beliefs stored in the short term memory to influence new beliefs. This is described in detail in Task 2.

Once we have the fused sensory data, we can draw meaningful conclusions (such as the identity, location and properties) about various objects present in the environment. This information is stored in an occupancy grid framework. The spatial belief manager is responsible for propagating these beliefs through the occupancy grid. The map manager overlays the occupancy grid with potentials, creating a potential gradient map. This potential map defines the gradient of the potential field associated with the entire map and can be used to perform path planning and path following. The map creation and management processes are described in detail in Task 3.

The current control algorithm uses dead reckoning and the potential gradient map to generate speed and steering commands for the robot. Dead reckoning is used to localize the robot and find its position in the map. The potential gradient at that location is used to generate the next set of navigation commands to the robot. This is described in Task 4.

On To Task 1.

Up to Main Page

This page is maintained by Amit Singhal.

Last update: 11/17/97.