ML Agent(tag)

Back

Seeker Agent

I started with making the Seeker Agent. His purpose is to tag the runner agent in a certain amount of time.

While working on it I encounter a problem. After it the completion of a training session an error occurred, and it didn't import the onnx file.

The onnx file is the brain file of the AI. You can implement the brain and start the scene without starting a training session. After downloading the right packages everything was solved.

Gif 1
Image 1
Gif 1
Image 2

Runner Agent

As next i made the Runner Agent. His purpose was to outlive the Seeker Agent until the timer reaches zero.

At the beginning of the trainingsessions i used a already trained onnx file of the seeker. So the runner had it twice as hard.

I decided that both agents should train together so that they both had an equal chance of winning.

Jump mechanic

After that I wanted that both agents could spring so they can use the environment in there advantage.

At first the agents would fall out of the map by bugging through the ground.

I fixed it to reduce the force (gravity/mass) it receives to prevent it from falling through the ground.

Gif 1
Image 2
Gif 1
Gif 2

Training Sessions Conclusion

The majority of my time is dedicated to developing, refining, and optimizing scripts. While continuous improvement and optimization are generally beneficial, they have proven less effective in this particular instance.

Due to insufficient training, the performance of both agents remains at a minimal level, a limitation that becomes evident when the ONNX file is activated. This underlines the importance of proper training to achieve meaningful results.

I will prioritize a more robust training phase. This approach will ensure that the models achieve a higher degree of accuracy and efficiency, aligning with the overarching goals of innovation and performance excellence.