PitchAI – Software Update Notes
We’ve listened to your feedback and assessed any errors to improve PitchAI! With our latest release, you may see some minor changes in numbers due to some major PitchAI improvements.
First, we have improved the 3D tracking model. This new model does a better job tracking throwing arm motion, in particular shoulder external rotation angle. We have tuned our data smoothing protocols, so the 3D model fully captures the rapid forearm motion, but still gives a smooth (not jerky) motion that aligns with the raw video data. The details beyond that are a closely guarded secret ?
Second, we have added a simulated ball position. PitchAI previously tracked the wrist of the throwing arm (you’ve probably noticed our mannikin has no hands), but now tracks a simulated ball position for the 2D overlay video and for the advanced 3D metrics (arm speed, arm path).
Third, we improved our torque prediction algorithms. Using a new inverse dynamics model based on gold standard biomechanics lab data, PitchAI now predicts torque based on a combination of posture and joint velocity metrics. While pitching elbow torque values show system-specific differences, (ie. Motus sleeve vs. lab torque calculation), average PitchAI torque values should be similar to what you see in the biomechanics literature.
Fourth, low frame rate videos should error out less frequently. Video at < 60 fps (typically 30 fps) would often fail in the previous build, but should run successfully in the new build. Note that we still recommend 120 fps+, as lower frame rates have too much arm motion blur and don’t produce great analysis results. We are working on a messaging system in the app to notify you of these data quality issues. For now, try to stick to uploading 120 fps+!
We have improved several other minor features and fixed a bullpen full of bugs. As pitches roll in, we monitor any issues and will fix them for the next release. Keep telling us how you’re using PitchAI, what you love about it, what could be improved, and we’ll keep getting our data closer to a lab-based mocap system!
