You are absolutely right. We mentioned in our last devlog that 0.5 will do exactly that. We will be expanding our training dataset to include helpers for specific scenes and trying out new AI network architectures and experimenting some more. Thanks for the feedback.