In the first versions of the benchmark, a transmission control protocol (TCP) interface for controllers was provided so that controllers written in other languages than Java could be interfaced to the code. However, this TCP interface introduced aconsiderable communication overhead making non-Java agentsorders of magnitude slower, andhad occasional stability issues.For later releases, a new library called A miCowas developed for communication between the benchmark and agents developedin other language.TheAmiColibrary is applicable for inter-language process communication beyond the Mario AI benchmark. The purpose of the library is to provide an easy-to-use and as seamless as possible bridge between foreign programming languages pre-serving high performance, comparable to the native languages’