Papers

6

Total Citations

178

H-Index

5

About

Martin Engelcke is a leading researcher at the intersection of generative modeling, computer vision, and robotics, with a core focus on object-centric scene understanding and robot manipulation. His most influential contribution is the **GENESIS** model (2019/2020, ~145 combined citations), a pioneering generative latent-variable framework that explicitly captures the compositional nature of visual scenes. By enabling unsupervised inference of object-centric latent representations, GENESIS laid critical groundwork for a new class of scene understanding models that treat visual scenes as collections of discrete objects rather than monolithic images. Building on this, Engelcke developed **APEX** (2021), which extends object-centric segmentation into the temporal domain for robot manipulation tasks, and **Reaching Through Latent Space** (2022), a novel path planning approach that optimizes robot trajectories directly in a learned latent space. His recent work on quadruped locomotion, including **VAE-Loco** (2023), demonstrates his versatility by applying disentangled representation learning to generate versatile, continuously variable gaits for dynamic robots. Engelcke’s research uniquely bridges unsupervised learning and practical robotics, enabling machines to perceive, segment, and interact with their environments in more human-like, compositional ways.

Research Focus

Key Achievements

5
H-Index
6
Papers
178
Total Citations
30
Avg Citations/Paper
🏆 Most Cited Paper
GENESIS: Generative Scene Inference and Sampling with Object-Centric Latent Representations
73 citations · 2019
📈 Most Prolific Year: 2022 (2 Papers)
🤝 Key Collaborators: 13
🏛 Institutions: University of Oxford, Robotics Research (United States), Google DeepMind (United Kingdom)

Top Papers

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6

Key Collaborators

Contact & Links

Available for collaboration
Content generated · 13 days ago