Recursive discovery. Real understanding.
Deep SIML Labs develops simulation-first systems for models, optimization, and scientific learning.
Our research pillars
Each pillar contributes to the same question: how can a research system learn more from every evaluation without overstating what has been demonstrated?
Adaptive memory, representation, and prediction error treated as testable mechanisms.
View direction →Model-driven environments for asking controlled computational questions.
View direction →Fair, budgeted comparison across complex design spaces.
View direction →Learned representations connected to constraints, uncertainty, and scientific models.
View direction →A long-term program for functional polymers, compliant systems, sensing, and actuation.
View direction →Current focus
The immediate work is to make the computational stages rigorous, reproducible, and comparable. Planned physical stages are shown honestly as future extensions.
The purpose of the loop is not to generate attractive predictions. It is to expose assumptions, quantify uncertainty, and make the next evaluation more informative.
* Physical capability planned / being built
Program status
A clear boundary between public computational work and the physical research capability still to be built.
Latest from the lab
Real project artifacts stay visible. Historical work stays available with its original context clearly marked.
A reported 100-episode simulation benchmark with method, code, and result figures.
A technical reading of V-JEPA 2 and its relevance to embodied simulation research.
An early embodied-agent artifact, retained with updated historical framing.