Erik Learned-Miller
Papers
7
Total Citations
43
H-Index
5
About
Erik Learned-Miller is a leading researcher in robotics and computer vision, whose work focuses on enabling autonomous systems to perceive, learn from, and interact with their environments. His key contributions span error detection in stochastic robot actions, affordance-based object modeling, and visuomotor learning. Learned-Miller pioneered the Aspect Transition Graph (ATG), a framework that models objects by their functional affordances—how they can be manipulated—allowing robots to plan complex manipulation sequences. His work on convolutional neural networks for grasping (2016) helped bridge deep learning and robotics, mapping visual features directly to pre-shape grasps for humanoid hands. More recently, he has advanced event camera-based visual odometry for agile-legged robots, achieving robust pose estimation during dynamic locomotion and acrobatics (2023). With over 43 citations across his most-cited papers, Learned-Miller’s research has influenced both theoretical models of object representation and practical systems for persistent backgrounding and error recovery. His work is essential reading for students and researchers interested in the intersection of perception, learning, and manipulation in robotics.
Research Focus
Key Achievements
Top Papers
- 1Error detection and surprise in stochastic robot actions10 citations · 2015
- 2
- 3
- 4The Aspect Transition Graph: An Affordance-Based Model6 citations · 2015
- 5Modeling Objects as Aspect Transition Graphs to Support Manipulation6 citations · 2017
- 6Associating Grasping with Convolutional Neural Network Features.3 citations · 2016
- 7