Max Robotics — Global Service Robot Directory + US Certification

About

Daniel Marta is a leading researcher at the intersection of robotics and artificial intelligence, whose work focuses on making deep reinforcement learning (RL) both safer and more aligned with human values. His primary contributions address two critical challenges: ensuring physical safety through formal verification and improving human-robot communication through preference-based learning. Marta’s most influential work, "Human-Feedback Shield Synthesis for Perceived Safety in Deep Reinforcement Learning" (20 citations), pioneered a novel approach that integrates human feedback into safety shields, avoiding the overly restrictive behaviors typical of adversarial assumptions. Building on this, his series of papers—including "Aligning Human Preferences with Baseline Objectives" and "VARIQuery" (13 citations each)—introduced innovative methods for efficiently querying humans to shape robot reward functions, dramatically reducing the burden of reward engineering. His recent work, "PREDILECT" (10 citations), advances zero-shot language-based reasoning to interpret human preferences, while "POLITE" (8 citations) combines preferences with visual highlights for more intuitive teaching. Marta’s research has earned him recognition for developing practical, human-centered RL systems that bridge the gap between theoretical safety guarantees and real-world social robot interactions, making him a rising star in the field of human-in-the-loop machine learning.