Computer generated holograms (CGHs) used for AR/VR displays typically have poor image quality compared to their numerical reconstructions. The primary cause of these discrepancies can be attributed to over idealized wave propagation used for CGH generation. This problem is often exacerbated when the image is projected through a more complex optical system like a holographic waveguide combiner or using the Texas Instruments phase light modulator (PLM) that has nonlinear phase levels. Direct camera feedback during hologram optimization has been shown to significantly improve image quality f or CGHs projected in f ree s pace. Here we demonstrate the use of camera feedback optimization for improving image quality of CGHs projected through holographic waveguide combiners and using the TI PLM as phase SLM. Machine learning can be applied to create adjust the numerical propagation method to closer match physical propagation through the system without the need for camera feedback after training. This method corrects for various optical aberrations, beam profile, and phase nonlinearities in the display. Further image improvement is made by leveraging high speed MEMS based PLM for time multiplexing CGHs to reduce speckle. Application of these techniques for different waveguide geometries (i.e. planar and curved) will be discussed.