Alberto Rivas

Google (United States), University of Toronto

Papers

Total Citations

H-Index

About

Alberto Rivas is a researcher at the forefront of integrating formal methods with robotic learning, with key contributions in vision-based manipulation and automated software synthesis. His work addresses fundamental challenges in computational efficiency for deep reinforcement learning, particularly in reward specification for robotic tasks. Rivas's highly cited paper "Reward Machines for Vision-Based Robotic Manipulation" (18 citations) pioneers a novel framework that reduces the computational complexity of Deep Q-Networks (DQN) in vision-guided manipulation, enabling robots to learn complex tasks with greater sample efficiency. This work bridges the gap between symbolic reasoning and continuous control, offering a principled approach to reward shaping. Additionally, his development of "SynKit: LTL Synthesis as a Service" (12 citations) makes Linear Temporal Logic synthesis accessible as a cloud-based tool, advancing the practical deployment of formal verification in software engineering. By combining theoretical rigor with applied robotics, Rivas has established himself as a key figure in the synthesis and learning communities, with his tools and methodologies directly influencing how autonomous systems reason about long-horizon tasks.