VAMPIRES – Virtual Agent Mission-Predictive Intelligent Response Engine for Space

We propose a solution framework to the problem in Space Traffic Management of unpredictable human-machine interactions during Conjunction Assessment (CA) / Collision Avoidance (COLA) situations.  Factors that make COLA involving autonomous space objects (SO’s) more complex include that SO’s may be uncontrolled, non-communicative, self-directing, and/or adversarial.  Versus COLA with a cooperative or uncontrolled SO, the COLA situation is more difficult when responding to a potential collision between a controlled, "friendly" asset and an unknown, potentially autonomous, potentially adversarial spacecraft. Our architecture will be designed to accommodate non-cooperative SO’s, whether disabled, non-communicative, self-directed, or adversarial. High-uncertainty scenarios and autonomous systems can be difficult to model sufficiently, and the goal is to guarantee proper, predictable behavior of the onboard autonomy, to ensure that mitigating COLA actions are in fact achievable, and that the avoidance maneuver will occur safely with high confidence. To do this, we propose a generational leap in autonomy based on: 1) collaborative space-ground networked agents, 2) autonomous onboard COLA, 3) Virtual Agent Autonomy to model intents of other intelligent agents via streaming contextual behavior modeling, 4) and next-generation safe autonomy algorithms using Shielded Deep Reinforcement Learning to prevent the autonomous spacecraft from entering into dangerous or impossible-to-recover situations. The combined technologies will be deployed on an integrated ground and flight software system which leverages existing SSA sensing data, SSA systems, and Defense and commercial space mission operations infrastructure. Initial checkouts of the system performed in a live-virtual constructive environment (LVCE), allowing simulation of the autonomy algorithms using operational data ingested in real-time to develop trust in the autonomy.