Prev-Talk | Next-Talk | All-Talks | Talks-Sorted
Talk-22: An Extensible Arbitration Framework for Predictive Modeling
Jean- François Kamath and Signe A. Redfield, NSWC-PCD
A useful technique for adaptive systems is forward projection that allows selection of more optimal solutions than local modeling alone can accomplish. This paper presents an approach for predictive modeling that enables easy implementation of decision making while reusing existing resources. The approach involves generation of dynamically generated simulation environments that execute identical software as their real-world counterparts using a modified data set. Not only does this ensure that the simulations will produce similar results to the actual system, but it also reduces the amount of development time required to implement or modify a predictive model. This paper also discusses an implementation of this framework in MOOS and some of the challenges involved in adding uncommon functionality.
Categories:
- Simulation
- Autonomy
- Govt Labs