Heron Systems was selected as one of eight companies/universities to participate in DARPA’s Alpha Dogfight Trials in August 2019. Using deep reinforcement learning, a novel training pipeline, and other innovations, we trained an AI agent that placed first in the three Trials held to date. The agent performs advanced tactical planning on long time horizons. It is able to operate within the latency requirements of the target system.
On 8/18 – 20, our agent will compete in the final Trial. The trial will feature a qualification round against DARPA provided agents and a round robin against other trialists. The top 4 teams will compete in a single elimination tournament to determine the ADT Champion and a chance to become the first AI to defeat a human-expert pilot. Stay tuned for updates.
Optimized training process able to more fully explore real-world conditions and stress the AI’s tactical understanding
Mature input interface able to adapt to numerous simulation environments
Experiment to production pipeline supporting constant innovation and seamless transition to production without interrupting training.
Refined training DEVOPS implementing real-time agent monitoring, fault tolerance, automated recovery, and hardware monitoring.