Marcell Vazquez-Chanlatte

University of California, Berkeley, Silicon Valley University

Papers

Total Citations

H-Index

About

Marcell Vazquez-Chanlatte is a leading researcher at the intersection of formal methods, robotics, and artificial intelligence, with a primary focus on learning interpretable task specifications from demonstrations. His foundational work, "Learning Task Specifications from Demonstrations" (2017, 42 citations), pioneered methods for inferring temporal logic specifications from human demonstrations, enabling robots to decompose complex behaviors into safe, composable sub-tasks with formal guarantees. This approach bridges the gap between intuitive human teaching and rigorous, verifiable robot control. Vazquez-Chanlatte further advanced the field with "Maximum Causal Entropy Specification Inference from Demonstrations" (2020), which introduced probabilistic frameworks for capturing temporal properties while maintaining safety guarantees. His recent work, "Diffusion-Based Failure Sampling for Evaluating Safety-Critical Autonomous Systems" (2025), tackles the critical challenge of validating autonomous systems in high-dimensional domains, offering a more efficient alternative to traditional Markov chain Monte Carlo methods. By combining formal verification with learning from demonstration, Vazquez-Chanlatte's research provides a principled pathway toward trustworthy autonomous systems that can learn complex, multi-stage tasks from human teachers while ensuring safety and correctness.