OmniScientist: An Omni-Modal Omni-Discipline AI Scientist

TL;DR - OmniScientist is an end-to-end, multimodal AI scientist that uses autonomous agents to turn heterogeneous raw evidence into research ideas, experiments, and compiled manuscripts. Direct perception significantly outperformed reliance on precomputed scalar features, highlighting its importance for evidence-grounded discovery.

• Combines a perception layer with agents for ideation, experimentation, and writing in a deterministic pipeline.
• Supports images, signals, audio, video, 3-D structures, trajectories, tables, formulas, and graphs across multiple disciplines.
• Code-based checks enforce novelty screening, statistical validity, provenance, and numerical traceability.
• Completed all 36 evaluated research cases; direct perception improved all seven evaluation dimensions and won 85% of paired judgments.