Rakhil Soman
Software Development
«
Machine Learning Based Real-Time Strategy AI
With the innovation of high-performance The current model is trained for more
available for machine learning, research rules, which lead to a selection of five
hardware, as well as it being more readily
is becoming much easier. Even games
are taking advantage of these systems in
many ways.
My objective with this project was
to create Real-Time Strategy (RTS)
different behaviors. After the training, RTS
AI’s winning percentage against humans
increased by 75 percent, whereas the
winning percentage against Random AI
increased by 100 percent.
game Artificial Intelligence (AI) using This proves that machine learning AI
RTS game AI architecture is developed currently existing AI architecture. With
machine learning techniques. Most of
using finite state machines or utility
based systems. In a traditional RTS
AI, system designers are required to
develop behaviors and mathematical
representation. Neural network based
RTS AI architecture enables the designer
to add and remove new behaviors with
minimum programming knowledge. These
models can also adapt and learn human
strategies by extracting the details of the
gameplay.
96
than 15 hours to achieve seven different
SOFTWARE DEVELOPMENT
can be a viable replacement for the
new advancement in hardware and
its performance capability, it will also
become easier to develop these systems.