Hybrid Deliberative-Behavior Based Reactive Control Implemented in a Heterogeneous Blimp Swarm
Abstract
This work will develop a hybrid swarm of heterogeneous, computationally lightweight sensing and compute vehicles to compete in an Of,fice of Naval Research (ONR) sponsored competition. The vehicles will be lighter-than-air blimps. The swarm is doubly hybrid because, 1) the blimps will not all be the same, and 2) the individual blimps will also contain hybrid control architectures (Deliberative-R,eactive). In addition to Artificial Intelligence (AI) techniques, behavior based autonomy architectures such as Subsumption will be, used so that we can incrementally build increasingly complex behaviors. Behavior specializations will include both offensive and d,efensive behaviors. An overriding insight is that the swarms population is important, not the individual. The heterogeneous blimp,team will be composed of specialized vehicles with primary behaviors, and coordination among the vehicles and the hybrid control arc,hitectures is the key to the approach. When communication is needed, non-Radio Frequency (RF) vehicle-to-vehicle communication will, be explored, e.g. light based, sound based, and tactile. Unconventional navigation and locomotion aids will also be explored. The, approach is special because we will find the boundary between Deliberative and Reactive control in the competitions problem, and s,ystematically work to move the boundary towards Reactive control. We will use the Subsumption architecture to build up behaviors th,at we develop and incrementally learn how to complete complex tasks. A hybrid Deliberative-Reactive control structure is useful bec,ause it uses Reactive control when appropriate and Deliberative control when appropriate. Our approach is also special because we w,ill take a minimalistic approach to the problem and seek to use not only digital, but analog processing. Reactive Sensing and Actin,g Behavior Based methods are in contrast to Deliberative Artificial Intelligence (AI) based methods. In Deliberative control method,s, an internal model of the environment is created, and that knowledge is used to drive robot planning and action. This internal mo,del, however, is not needed in Behavior Based approaches where The world is its own best model, and this is special. Although the,re are a number of behavior based approaches, we will initially focus on Subsumption which allows for incrementally learning in a,Build-Test-Build environment. Its architecture can be contrasted with the traditional AI Deliberative method of sense-plan-act.The,projects objective is to develop a hybrid deliberative/reactive autonomy architecture that is effective in a dynamic environment wh,ere the environment also contains unstructured aspects. This autonomy architecture will empower our mobile, lighter-than-air robots, to score points in a competition. This scoring is in spite of a competitor who is also trying to score and possibly trying to deny, our team from scoring. The expected performance improvement of our autonomy architecture is that it will be more robust than other, architectures to changes in the environment and to unknown actions of our competitors.This proposals Naval relevance is related to, the autonomy and adversarial aspects of this project. This research will develop robust autonomous systems for operation in dynami,c and unknown environments, and this has direct relevance to operational environments and situations that Navy autonomous systems en,counter.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 16, 2022
- Source ID
- N000142212190
Entities
People
- Scott Koziol
Organizations
- Baylor University
- Office of Naval Research
- United States Navy