ROLE: Researcher / Robotic Systems Engineer

SKILLS UTILIZED: Systems Design, Coding (C++, Python), Data Analysis

SOFTWARE: ROS, ROS 2, Bebop_Autonomy SDK, Tensorflow, Sonnet, HBP Neurorobotics Platform, OpenAI Stable Baselines


HARDWARE: Parrot Bebop 2


Current applications of machine learning in multi-robot systems (MRSs) isolate the learning of robots and do not capitalize on the presence of multiple agents. Once models are deployed, there is no sharing of knowledge between agents, thus learning processes are restricted to the knowledge of a single robot. We propose Unified Learning, an MRS architecture that optimizes machine learning algorithm performance through the unification of individual robot knowledge to leverage the power of the crowd. The architecture is inspired by shared mental models theory from cognitive psychology, which is proven to both describe and promote optimal team performance and communication. The results of an implementation of Unified Learning with multiple UAVs demonstrates that by applying shared mental models to MRSs, robots will develop knowledge more quickly and efficiently, thus optimizing system performance. 

Unified Learning

Multi-Robot Machine Learning using Shared Mental Models (Preprint)