Type:        Internship
Employer: Heron Systems, Inc. (Shield AI)
Program:  Defense Advanced Research Projects Agency
                 (DARPA) -
Air Combat Evolution
Duration:  May 2021 - August 2021
Designed and implemented a new multi-agent reinforcement learning framework to scale up and extend the winning AI pilot of the AlphaDogfight Trials, with a focus on flexibility and extensibility to enable its control of multiple, heterogeneous collaborative combat aircraft;
Trained prototype agents for 2-on-2 aerial dogfights by shaping rewards for both combat performance and flight safety, initiating the work to induce complex collaborative behavior;
Supported Heron Systems’ success in proceeding to Phase 2 of the program in a down-select from eight to three contractors, cutting Lockheed Martin and Boeing.