I implemented some new behaviors for the AI and switched from the 2d grid to a much more nav-mesh-like representation. Since I still wanted to use the same data structure for both pathfinding and influence mapping, the graph has to be rather dense. But performance on the CPU-side is not really a problem, so I'm okay with it. I also had a look at Dave Mark's and Kevin Dill's talk on utility-based AI and promptly implemented some of the concepts.
The shader for the skill that will essentially be the robot version of night vision turned out well, I think. I always thought the visuals of the night vision modes in other games were way too clean, so I put a lot of noise and blur in it. It looks much more coherent in motion though. Our brains are way too good at recognizing shapes. (Same scene with normal shaders).
I love tinkering with a shader until it looks nice and I'm fairly okay at creating textures but I was definitely not born to be a 3d artist. Since everybody seems to love cubes nowadays anyway, I just decided to embrace a more abstract and simplistic style for now. We'll see how that turns out