Emergent Awarded NASA Phase 1 SBIR Contracts for Autonomous FSW

Emergent Awarded NASA SBIR Contracts for Small Satellite Autonomy Flight Software

Emergent Space Technologies, Inc. (Emergent) was awarded two contracts by NASA under its Small Business Innovation Research (SBIR) / Small Business Technology Transfer Research (STTR) program to prototype flight software for autonomous satellites. NASA’s SBIR/STTR program funds the research, development, and demonstration of innovative technologies that fulfill NASA needs as described in annual solicitations and have significant potential for successful commercialization. The artificial intelligence (AI) and machine learning (ML) domains contribute to the proposed innovations.

The “Plan Generation for Autonomous Small Spacecraft Swarms” project will prototype planning software for autonomous vehicles, which includes but is not limited to small satellites (SmallSats). Swarms and constellations of SmallSats are being increasingly used in low Earth orbit (LEO), and government operators have expressed interest in missions at higher altitudes, e.g. GEO, as well as for proximity operations. These new mission regimes are more challenging and require greater onboard autonomy capabilities to act and plan during safely-critical mission phases. Planning software enables missions with large numbers of cooperating platforms to generate new plans as needed to respond to changing conditions.

An autonomous SmallSat swarm is collecting data according to a predetermined schedule and formation. When one vehicle experiences a failure, the fleet is able to re-plan to compensate for its absence. The replanning process occurs onboard the satellites and is aware of common sources of uncertainty.

Emergent’s proposed solution is a “safety aware” planner. The planner accepts developer-provided information about possible exceptions and incorporates information about recovery in the resulting plan. The planner should be able to execute in situ on the autonomous platform(s) to remediate the limitations associated with remote operation of small satellites. In addition, the planner is designed to be compatible with existing applications whose output can inform a plan for an overall mission. These capabilities make the planner framework very flexible and suitable for deployment on a host of candidate missions. Planning software for SmallSat draws from a large range of topics in traditional AI.

For solutions to planning problems, both sequential decision problems, typically represented by Markov Decision Processes (MPDs), and as well as efforts in adversarial search/game theory address the question of how an agent can achieve a goal in complex or partially known environments. Research in the area of probabilistic reasoning, such as Bayesian Networks, offers methods for an agent to represent and make inferences about uncertain events. Emergent will leverage our expertise in the areas of AI and ML in implementing our proposed planning solution. “Our goal is to generate plans that exploit the typical design of flight software applications,” said Dr. Timothy Woodbury, Emergent’s Principal Investigator. “This reduces the human effort needed to translate between the planning problem and the software that flies on the vehicles.”

When combined with cloud-computing software and hardware, ML will enable computationally intensive science data processing applications to be run on-orbit as distributed tasks.

The Machine Learning-Accelerated Grid Environment (MAGE) project is prototyping a software framework and application programming interface (API) that facilitates ML training and inference distributed across a networked constellation or swarm of SmallSats to enable resource intensive ML models to run at the extreme edge. This is accomplished by running on AI-accelerated hardware and distributing ML processing and storage across a grid of compute and storage nodes. Utilizing Emergent’s existing space-based cloud computing platform, Cirrus, as the foundation for MAGE, ML engineers can train and test models and run inference with state-of-the-art ML models in orbit instead of on the ground. Cirrus was prototyped under a prior U.S. Air Force SBIR contract to enable the use of SmallSat networks for on-orbit, cloud constrained small satellite environment. Cirrus is composed of Network, Storage, Compute, Sensor, Asset Scheduler, and Cyber services.

MAGE enables the proliferation of autonomous, intelligent systems to make informed decisions in the space domain. By providing a platform for ML interference in space, MAGE will enable a new generation of autonomous spacecraft that can gather information about their state, perform ML interference based on that state, and decide about how to respond. “Machine Learning is being increasingly utilized in the ground-based applications and there is ongoing research for on-orbit applications,” said Brett Carver, Emergent’s Principal Investigator. “The difficulty with running state-of-the-art machine learning on-orbit is overcoming the resource constraints. MAGE is one potential answer to this resource problem.” A possible future application would be creating an on-orbit system that is capable of discerning which data is significant enough to downlink, discarding uninteresting data, or a system.