As part of Shield AI’s work on an AFWERX Strategic Funding Increase (STRATFI) contract, we’ve integrated our autonomy and artificial intelligence stack, Hivemind, onto our vertical takeoff and landing (VTOL) unmanned aircraft, V-BAT, in simulation.
The goal of this project is to operationalize Hivemind for different hardware platforms to support Air Force Special Operations Command (AFSOC), Air Force Research Laboratory (AFRL), AFWERX, and Agility Prime. With this effort, we are flipping the paradigm of multiple warfighters supporting a single UAS to multiple UAS being commanded by a minimal cadre of warfighters, thus rapidly scaling the combat power, autonomy, and mass provided by Hivemind.
Hivemind can be trained to execute any number of missions to include integrated air defense breach, SCUD missile hunting, zone reconnaissance, counter-air, beyond-visual-range strike, maritime domain awareness, and communications-contested operations. These missions are built upon the same Hivemind architecture and simply require new models and scenario-generated inputs. For this project, we demonstrated Hivemind’s AI capabilities with a team of three V-BATs tasked with monitoring a wildfire in a 10 x 10 km region containing three smaller Named Areas of Interest (NAIs).
How Do Simulations Work?
The Shield AI Simulator is a proprietary high-fidelity, high-performance, scalable simulation environment that supports development, testing and evaluation of various Hivemind capabilities, including: sensor and vehicle characterization and modeling; environment modeling, automated testing, verification, and validation of algorithms; visualizations, user interfaces and experience for developers and operators, playback of synthetic and real-world missions; real-time command and control of operations; and the creation and consumption of training content via conventional and immersive interfaces.
We were able to complete multiple simulation iterations under various conditions, including the possibility of enemy threats or friendly forces being positioned in differing locations in the area.
Hivemind is an ecosystem of tightly integrated AI software tools and interfaces that provides best-in-class autonomy for robots, seamless interactions for their users and rapid velocity for their developers. This ecosystem spans software for the edge (“Hivemind Edge”), for the local and cloud-based development and analysis environment (“Hivemind Design”) and for the end user and their trainers (“Hivemind Commander”). Together, the components within this ecosystem strive to deliver intelligent and resilient autonomy capabilities for heterogenous teams of crewed-and-uncrewed aerial systems.
Hivemind supports multiple platforms via its modular open system architecture. Notably, it provides autonomous tasks and behaviors for single- or multi-agent systems through a catalog of re-composable “blocks” that can be combined to leverage the capabilities of a vehicle with payloads to execute diverse mission sets.
Hivemind is comprised of a group of software products used across programs and solutions. It consists of three top-level pillars, each of which are composed of multiple software product components. During this project, all three pillars of Hivemind worked together to create the simulation and autonomy that was demonstrated.
- Hivemind Edge: Software for the edge, designed to deliver intelligent autonomy and human-machine-teaming for a diversity of complex mission objectives within diverse data-denied, GPS-denied environments. Hivemind Edge leverages an open and modular architecture that is platform agnostic and designed to integrate with 3rd party systems. Hivemind Edge develops the foundations (hardware and platform abstraction layers, middleware) and higher-level AI capabilities (perception, action, cognition) for edge systems.
- Hivemind Design: Software for AI-developers and designers, intended to accelerate capability development and testing for autonomous systems using highly customized and optimized local and cloud compute resources. This group develops services for AI capability design, configuration, data management, analysis, testing, and reporting.
- Hivemind Commander: Software and hardware to support rich human-in-the-loop and human-on-the-loop interactions with the Hivemind Ecosystem that elevates human operators to mastery, while enhancing AI performance through simulated-and-real-world interactions with people. Hivemind Commander develops interfaces and methodologies for effective mission planning, command and control, training, and considerations of human factors across Hivemind.
The results of this project showcase the power of Hivemind to enhance the effectiveness of military operations. These enhancements will further develop our technology that can be easily and effectively employed by the warfighter. Follow-on milestones for the integration of Hivemind on V-BAT include several successful flight tests and the delivery of three prototype V-BATs, with Hivemind integrated, by this time next year.