Carlota Parés-Morlans
Carlota Parés-Morlans is a PhD student at Stanford University affiliated with the Knight-Hennessy Scholars program and the Institute for Human-Centered Artificial Intelligence (HAI). She is known for her research on vision-language-action models and contact-rich robotic manipulation, including co-authoring work on the FACT methodology. She holds a bachelor's degree from La Salle in Barcelona, Spain and is a recipient of a "La Caixa" Foundation fellowship.
“The paper's core contribution is a principled decomposition of why VLA models fail at contact-rich tasks into two distinct root causes: precision failures and force failures”
Source→“Across five contact-rich tasks and ~2,500 real-world rollouts, FACT achieves 66% average success vs. 41% for ForceVLA and 37.5% for TA-VLA (Table 1)”
Source→“The Logit-Normal (LN) noise schedule is a drop-in replacement for the Beta schedule used in π0, π0.5, and SmolVLA that reallocates 6× more gradient signal to the contact-correction regime (τ < 0.2)”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.