I recently tried the grid sensor and my agents has improved significantly.
Unfortunately for some reason (tensorflow limitations?) we’re forced to use a 20 by 20 grid resolution, which seems excessive.
Even just a few of these agents are unable to run smoothly on mobile devices, which my old ray sensor based agents were able to do without any issues.
Any way i can improve the grid sensor performance?
I had a work-in-progress pull request to Jobify the grid sensor but never finished it due to other high-priority tasks. You could try to take it forward. There were significant performance boosts, but some limitations came with them. [WIP] Jobify grid sensor by surfnerd · Pull Request #5021 · Unity-Technologies/ml-agents · GitHub
MLAgents Grid Sensor performance varies with the complexity of the environment. It’s efficient for simple grids but may lag with larger, more detailed grids, impacting training time and accuracy.
HQL Solutions - Best B2B Lead Generation Company