PROGRAMMING AND ART
This week, we focused on improving the quality of the animations which are recreated. The machine learning pipeline gives output as animation data (bvh) file, which is meant for a specific Blender skeleton. We realized that we were trying to solve a very difficult problem, by trying to re-target animation data in Unreal Engine to a differently rigged skeleton. This is complicated by the fact that the axis orientations of Unreal and bvh file are different. The table below shows this comparison.
This was leading to poor recreation of animations in Unreal Engine. To overcome this problem, we re-rigged our characters to exactly match the required rig. This led to a much simpler re-targeting process and significantly improved the quality of the animations.
On Wednesday (April 24, 2019) we also had a soft opening for our project. Guests from Electronic Arts visited our space and tried our experience. We got good reactions and feedback from the players regarding the experience. Here’s a video of our soft opening presentation
Some important takeaways from soft opening
- The players understood the instruction video and it also helped them get a sense of what they are getting into
- The players enjoyed recording videos for the abstract prompts we provided
- They liked the pay-off at the end, but hoped it could be an instant pay-off (instead of waiting for 20 minutes)
- We used a default skeleton model as our character during soft opening, and one feedback we got is that having our final character would help the game look more polished
- The players were not frustrated by the fact that the animations recreated were not high-fidelity animations, because the prompts set expectations for a silly/humorous experience
Next week, we will focus on polishing our documentation and product for delivery. We are also two weeks away from our final presentation, and will start preparing for it this week.