• Uses Machine Learning by Evolving Executable Programs that executes on Virtual CPUs; this is a form of supervised learning.
  • Demo evolves programs of robotic agents involved in a resource detection, extraction, and gathering problem in a dynamic, real-time, asynchronous environment. The robotic agents go from not knowing how to do anything to learning how to orchestrate their actions to start gathering resources efficiently.
  • Framework is extensible to generalize on any Virtual CPU specification with custom Instruction set.
  • Implements 3 evolutionary models:
    • Population Islands Model
    • Embodied Evolution Model
    • Artificial Life Model
  • Implements 4 interaction mechanisms:
    • No interaction
    • Trail (Stigmergy)
    • Broadcast
    • Unicast (closest Agent)
  • Implements 2 diversity modes:
    • Homogenous multi-function agents
    • Heterogenous single-function agents
  • Developed in MASON:  a discrete event multi-agent simulation platform
  • Agents are evolved using Genetic Programming using a custom extension of the Push language
  • Simulation designed to be distributed across several processors in the High Performance Compute Cluster (HPCC) at CUNY (College of Staten Island)