Herke van Hoof

About

Herke van Hoof is a robotics and machine learning researcher whose work sits at the intersection of reinforcement learning, tactile sensing, and dexterous robotic manipulation. His research addresses one of robotics' most persistent challenges: enabling robots to skillfully handle unknown objects without relying on pre-built models. By exploiting compliance and tactile feedback, van Hoof has developed methods that allow robots to learn in-hand manipulation, predict and prevent grip slip, and actively explore object surfaces using Gaussian processes — work that collectively spans over 500 citations across his most influential papers. A recurring theme in his research is making learning tractable from high-dimensional sensory input. His 2016 work on stable reinforcement learning with autoencoders (142 citations) demonstrated how robots can acquire tactile and visual skills through trial and error, while his hierarchical skill learning approach (109 citations) tackled multi-phase manipulation by decomposing complex tasks into learnable sub-sequences. Beyond manipulation, van Hoof has contributed to probabilistic scene exploration, LiDAR scan synthesis for robot mapping, and navigation via hierarchical reinforcement learning. His body of work reflects a consistent ambition: building autonomous robots capable of learning robustly from rich sensory experience in unstructured, real-world environments.

Research Focus

Key Achievements

Top Papers

1. 1
   Learning robot in-hand manipulation with tactile features

   165 citations · 2015
2. 2
   Stable reinforcement learning with autoencoders for tactile and visual data

   142 citations · 2016
3. 3
   Towards learning hierarchical skills for multi-phase manipulation tasks

   109 citations · 2015
4. 4
   Stabilizing novel objects by learning to predict tactile slip

   105 citations · 2015
5. 5
   Active tactile object exploration with Gaussian processes

   96 citations · 2016
6. 6
   Probabilistic Segmentation and Targeted Exploration of Objects in Cluttered Environments

   71 citations · 2014
7. 7
   60 citations
8. 8
   Maximally informative interaction learning for scene exploration

   44 citations · 2012
9. 9
   Non-parametric policy search with limited information loss

   24 citations · 2017
10. 10
    21 citations

Key Collaborators

Contact & Links