Marine Systems Engineering Laboratory:

EAVE Software Architecture

Over the past fifteen years, MSEL has developed and refined a hierarchical layered architecture for controlling AUVs, called the EAVE software architecture. The basic idea behind this architecture is to allow layers closest to the hardware to respond fast, and those further from the hardware to respond more intelligently. The latter includes pre-setting the lower levels so that they respond appropriately as well. The architecture is used to control the EAVE-III vehicles and will be used for most other vehicles MSEL develops.

The architecture has several modules:

• Sensor and Effector Managers: These handle the very lowest-level interactions with the hardware and will be specific to each different kind of vehicle on which the architecture is used.
• Data Assessment: This module takes the filtered sensor output of the Sensor Manager and converts it into objective, symbolic assessments of the world and vehicle (e.g., "depth is 5 meters").
• Situation Assessment: This takes Data Assessment's output and converts it into subjective assessments relative to the vehicle (e.g., "too deep").
• Mission Assessment: This takes the situation assessment and evaluates that to arrive at a current assessment relative to the mission in progress (e.g., "goal G1 failing").
• Mission Planner: This takes high-level, abstract goals and constraints from the user and creates a plan to accomplish the mission. It executes the plan by sending commands down to Navigator.
• Navigator: Navigator is responsible for fairly large-grain effector and motion commands. Path planning, for example, is at this level. Its tasks are accomplished by sending commands to Guidance.
• Guidance: The Guidance module is responsible for low-level motion and effector commands (e.g., "move to x, y, z"). This is the module in which obstacle avoidance algorithms reside.

New Directions:

The topmost level of the EAVE architecture has not been filled in. Currently, work the Orca project is remedying this situation. It is a schema-based, context-sensitive reasoner currently under construction at UNH and MSEL. As can be seen in the figure, Orca will replace both top-level modules.