Using Transfer Learning in an Ad Hoc Team

Abstract

In a practical scenario, we have a myriad of robotic systems; single and multi agent, not operating under any standard communication protocols. This is eminent from a point of view where independent robots can come together to achieve goals as a team. This problem is well defined in the domain of Ad Hoc Teamwork(AHT) which strives for a MAS wherein agents are heterogeneous, independent in their own respect and, as a whole accomplish goals that may be above any individual's capability. A key aspect is transfer learning which allows on the fly addition of team members. This paper shares development in transfer learning in the field of Ad Hoc Teamwork.