Probabilistic Technologies for Embedded Space Systems

A spacecraft payload in a hostile and threatening space environment requires a real time protective spacecraft response. Operationally responsive timelines which include telemetry downlink, collaborative analysis, and protective uplink commands to a spacecraft lack the timeliness required to successfully react to the threat. The strategy proposed in this paper advocates greater spacecraft autonomy. The challenge becomes successfully confronting the highly dynamic and uncertain spacecraft environment. Unanticipated issues always arise. Consequently, rule-based solutions lack the flexibility to be appropriately responsive. Our approach embeds a general purpose evidence fusion engine in the spacecraft to provide limited on-orbit autonomy to endangered spacecraft. Progress includes software prototyping of the processing thread and definition of mission use cases. Simulation of an orbiting anti-satellite threat and a defensive vehicle provides context for a defensive maneuver based on evidential reasoning algorithms. We also performed computer hardware implementation trades for a space-based embedded instantiation based on algorithm and data processing requirements. Introduction: Our goal focuses on a speedier spacecraft response to hostile space environments. The requirement for a real time response occurs when space weather, orbital debris, unintentional radio frequency interference, or hostile actions endanger a space payload. Timelines for operationally responsive spacecraft to accommodate mission health and status data downlink to a ground station, collaborative analysis, and command uplink to the spacecraft lack adequate span. The solution requires greater spacecraft autonomy. Associated with this solution is the challenge of providing evidence of a potentially dangerous situation that is sparse, incomplete, uncertain, and changes rapidly. Consequently, autonomy based on rules, command tables, and other forms of simple processing logic are inadequate. Our approach embeds a general purpose evidence fusion engine in the spacecraft to provide limited on-orbit autonomy to a spacecraft in danger. The success of the Stanford Racing Team winning the 2005 DARPA Grand Challenge for driverless cars credits probabilistic reasoning to its victory. In this event, vehicles passed through three narrow tunnels and negotiated more than 100 sharp left and right turns. The race concluded through a winding mountain pass with sheer drop-offs on both sides. Uncertainties associated with multiple and fairly poor on-board sensors were overcome by fusing the data to produce higher-accuracy position and orientation information. Updating vehicle status occurred by combining a-priori information on vehicle dynamics and the environment. Our technique for probabilistic reasoning 1 http://www.darpa.mil/GRANDCHALLENGE/Teams/Stanfordracing.asp, accessed 7 July 2009. AIAA SPACE 2009 Conference & Exposition 14 17 September 2009, Pasadena, California AIAA 2009-6811 Copyright © 2009 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved. employs a patented general purpose evidence fusion engine (GPEFE). This processing software ingests evidence that may be incomplete, vague, time-varying, uncertain, and possibly conflicting. Unlike Bayesian belief networks which scale exponentially and are therefore only useful for relatively small networks, our technology scales linearly in the number of nodes, links, time steps, and pieces of evidence fused. It allows us to accept structured or unstructured data, mathematically fuse it at nodes representing hypotheses, and propagate it to other nodes via links that represent the influence that one node has on another. This provides data fusion at all Joint Director of Laboratories levels: • Object (Level 1) • Situation (Level 2) • Mission (Level 3) • Refinement (Level 4) The latter challenge, which has future possibilities, is met by combining data fusion and data mining inferences to discover new nodes and links. Applications of an evidence fusion engine embedded in a spacecraft are: • Debris avoidance • Coordination among fractionated flyers • Hunter satellites • Avoidance of hostile threats such as direct ascent anti-satellite (ASAT) weapons, coorbital anti-satellites, and ground-based high power lasers. Technology for autonomous probabilistic reasoning consists of an embedded processing thread. Embedded algorithms ingest structured evidence from onboard sensors, other ground or space-based sensors, and alerts from command centers. Evidence is associated with hypotheses, given degrees of belief, ignorance, and disbelief, and fused with existing evidence. Evidence propagates from the evidence ingest layer to a task level: find, fix, track, target, act, and assess probabilities for the potential danger. Evidence then propagates to the outcome level to reason about probabilities of emerging dangers, actionable situations, and neutralized threats. A Best Next Observation algorithm and associated metrics to determine response options serve to evaluate hypotheses. These hypotheses link to dynamic response options, or case-based repairs, that are stored, updated from the ground station, and prioritized based on a learned value of predicted effectiveness. Progress includes software prototyping of the processing thread (Levels 1 – 3 Data Fusion) and definition of mission use cases. Simulation of an orbiting anti-satellite threat and a weather spacecraft, both in sun-synchronous orbits, provides context for a defensive maneuver based on evidential reasoning algorithms. We also performed computer hardware implementation trades for a space-based embedded instantiation based on algorithm and data processing requirements. Mission Conditions and Constraints: 2 http://patft.uspto.gov/netacgi/nphParser?Sect1=PTO1S e.g., the BAE RAD750 , is similar to the earliest Pentium processors from the mid 1990’s with 266 MIPS performance. This practical limitation and on board memory (e.g., 44MB for the BAE RAD750) restricts the amount of real-time processing that can be conducted with expanding to multi-computer deployments Belief Network Scalability: Since embedded space computing environments are extremely limited compared to modern ground computing environments, it is important to evaluate the processing requirements of a General Purpose Evidence Fusion Engine (GPEFE) as an early part of the tradeoff analysis. Our initial qualitative analysis shows that memory is the limiting factor on the number of nodes and links (~60 Kb per node with 3 links/node) in a GPEFE and the processor speed being the dominate factor in belief updates frequencies. For large networks, the low belief update frequency limitation due to processor speed becomes the largest impediment deployment as the reduction in latency becomes inconsequential compared to ground processed architectures. Novel approaches have been investigated to lower the computational burden in embedded environments