Robot Intelligence Kernel
Architecture
The INL's generic, object-oriented software architecture has provided intelligence to a variety of today's commercially available robot systems. The unique object-oriented framework of the overall software architecture coupled with the use of "perceptual abstractions" allows the system to be easily reconfigured for new robot platforms and sensor suites. Unlike other attempts to develop robot intelligence and behavior that have been platform and domain specific, RIK can be reconfigured without the need to rewrite behavior modules. Without a single source code change, the same RIK software package can be run on vastly different robots, varying in size from several inches to several meters in length. Once a robot is defined within RIK it can be configured with any combination of perceptual payloads available. Another advantage to this plug-and-play reconfigurability is that it can adapt the behavior of the robot to real-time failures of actuators and sensors. The two diagrams below show how raw information flows and is processed throughout the RIK.
(See full-size image)
The first diagram below shows how sensor data and robot state is abstracted into the building blocks of behavior.
(See full-size image)
After the raw data is fused into these abstractions, the RIK further processes these abstractions by filtering each of them based on timing and significance.
(See full-size image)
The next diagram shows how the resulting abstractions are then used by a suite of interconnected behaviors. The cognitive glue within the RIK orchestrates these behaviors to accomplish various missions and competencies:
(See full-size image)
