Max Robotics — Global Service Robot Directory + US Certification

About

Joey Hejna is a leading researcher at the intersection of robotics, reinforcement learning (RL), and human feedback, whose work is helping to define how robots learn from people. Hejna’s key contributions span two critical areas: developing large-scale, open-source datasets and algorithms for generalist robot manipulation, and pioneering new methods for aligning robot behavior with human intent without relying on handcrafted reward functions. His most impactful work, the DROID dataset (108 citations), is a large-scale, in-the-wild robot manipulation dataset that serves as a foundational resource for training robust, generalizable policies. Building on this, Hejna co-authored Octo, an open-source generalist robot policy (66 citations), which enables pretrained models to be finetuned with minimal in-domain data. In the realm of human-aligned RL, Hejna introduced Inverse Preference Learning and Contrastive Preference Learning, which elegantly bypass the traditional reward model bottleneck by learning directly from human preferences—a significant leap in efficiency and safety. His work on Few-Shot Preference Learning further empowers human-in-the-loop systems, making it practical to teach robots complex tasks with minimal feedback. Through these contributions, Hejna is making robots that are both more capable and more intuitive to teach.