S2S: Physics Certification for Physical AI Training Data

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Physical AI systems — robots, prosthetics, and embodied agents — train on sensor data (IMU, EMG, PPG) that is routinely assumed to be clean. S2S is a physics certification engine that evaluates each sensor window against 16 biomechanical laws and assigns a quality tier (GOLD, SILVER, BRONZE, REJECTED). Validated on five public datasets. On PAMAP2, removing physics-rejected windows improves activity recognition F1 by +4.30% using 22.7% less training data. The Physical Action Tokenizer (PAT) demonstrates that certification tier predicts action uncertainty with 18.8x discrimination (REJECTED/GOLD entropy ratio). Code: pip install s2s-certify