So after changing the reward math again and tweaking the train script we got a big jump, i trained the model for 1lack steps and it came to 3k frames, my next goal is to increase the difficulty of the environment to make it challenging for the model