James MacGlashan
Brown University, Laboratoire d'Informatique de Paris-Nord, John Brown University
Papers
13
Total Citations
506
H-Index
11
About
James MacGlashan is a leading researcher in interactive robot learning, human-robot interaction, and reinforcement learning from human feedback. His work fundamentally addresses how robots can learn complex behaviors and tasks through natural, intuitive communication with non-expert humans—without requiring programming knowledge. MacGlashan’s major contributions include pioneering methods for interactive learning from policy-dependent human feedback, where a robot learns from a human teacher’s positive and negative evaluations that are themselves influenced by the robot’s current behavior. He also developed techniques for grounding natural language commands directly into reward functions, enabling robots to understand and execute spoken instructions. His highly cited papers—including “Interactive Learning from Policy-Dependent Human Feedback” (108 citations) and “Social is special: A normative framework for teaching with and learning from evaluative feedback” (91 citations)—demonstrate significant impact. MacGlashan’s work on reducing errors in object-fetching through social feedback and planning with abstract Markov decision processes has advanced the practical deployment of social robots. He is also recognized for using Minecraft as an experimental testbed for AI and robotics research.
Research Focus
Key Achievements
Top Papers
- 1Interactive Learning from Policy-Dependent Human Feedback108 citations · 2017
- 2
- 3Grounding English Commands to Reward Functions65 citations · 2015
- 4Reducing errors in object-fetching interactions through social feedback65 citations · 2017
- 5Planning with Abstract Markov Decision Processes49 citations · 2017
- 6
- 7
- 8Goal-Based Action Priors23 citations · 2015
- 9Minecraft as an Experimental World for AI in Robotics.17 citations · 2015
- 10Training an Agent to Ground Commands with Reward and Punishment12 citations · 2014