Max Robotics — Global Service Robot Directory + US Certification

About

Gabriel Dulac-Arnold is a prominent researcher working at the intersection of robotics, reinforcement learning, and multimodal AI systems. His work spans several interconnected domains, including offline reinforcement learning, robotic manipulation, legged locomotion, and vision-language models for embodied reasoning. Among his most significant contributions is the RoboVQA framework, which introduces a scalable, bottom-up data collection scheme enabling long-horizon multimodal reasoning for robotics — achieving 2.2x higher throughput than traditional approaches and garnering 34 citations since its 2024 publication. His work on Model-Based Offline Planning (2020) addresses a critical challenge in real-world RL deployment: learning effective policies without direct system access, accumulating 20 citations. He has also advanced reward learning from human demonstration videos, offering a cost-effective bridge between human behavior and robotic manipulation skills. Dulac-Arnold's contributions extend to benchmarking agile quadruped locomotion through the Barkour benchmark, and to residual reinforcement learning from demonstrations, reflecting his commitment to grounding theoretical advances in practical robotic systems. Collectively, his research portfolio demonstrates a sustained effort to make reinforcement learning and robotics more scalable, data-efficient, and applicable to complex real-world environments.