Karol Hausman

Researcher in robotics
and machine learning

About

I'm a PhD student at the University of Southern California in the Robotics Embedded Systems Lab (RESL) under supervision of Prof. Gaurav Sukhatme. I'm also closely collaborating with Stefan Schaal's group at USC. I'm mostly involved in interactive perception, reinforcement learning and probabilistic state estimation, however, I have very broad interests in the fields of robotics and machine learning. While being at RESL I did a number of internships at: Bosch LLC (2013 and 2014) working on active articulation model estimation, NASA JPL (2015) working on multi-sensor fusion and Qualcomm Research (2016) working on active mapping and planning under uncertainty. In summer 2017 I will be joining Google DeepMind for another exciting internship.

My research interests lie in active state estimation, control generation and machine learning for robotics. I investigate interactive perception, by which robots use their manipulation capabilities to gain the most useful perceptual information to model the world and inform intelligent decision making. The paradigm of generating motion to improve state estimation (interactive perception) and task execution (reinforcement learning) is applied throughout my work, in which I show that coupling perception and control together can be beneficial for both fields. More recently, I have been investigating deep reinforcement learning and its applications in robotics. I have evaluated my work on many different platforms including quadrotors, humanoid robots and robotic arms.

I studied Mechatronics and Computer Science at Warsaw University of Technology (WUT), Philosophy at the University of Warsaw and Robotics, Cognition and Intelligence at Technical University of Munich (TUM). In August 2012, I graduated summa cum laude obtaining Master of Robotics degree at WUT. In December 2013, I graduated summa cum laude with a second Master of Computer Science degree from TUM.

News

  • I'm happy to announce that I will be spending Summer 2017 at Google DeepMind in London.

  • I'm excited to be in the Program Committee of the RSS 2017 Workshop: Revisiting Contact - Turning a problem into a solution. Please contribute!

  • In April 2016, I passed my PhD qualification exam a.k.a. thesis proposal.

  • I co-organized the RSS 2016 Workshop on Robot Environment Interaction for Perception and Manipulation. Please see the details here.

  • We released our BioTac Grasp Stability Dataset (BiGS)! If you always wanted to do interesting machine learning on complex tactile data, you should try it out! You can find it here.

  • I gave an invited talk on “Multi-Sensor Fusion with Seamless Sensor Switching and Trajectory Optimization for Self-Calibration” at Google Tango (10.2016), UCLA (10.2016) and Qualcomm (06.2016). Thanks for having me!

  • I gave an invited talk on “Active and Interactive Perception” at Stanford (10.2016) and NASA JPL (09.2015). Thank you for the invitations!

Research




(Deep) Reinforcement Learning for robotics

In my research on Deep Reinforcement Learning, I focus on how we can apply modern deep learning techniques while leveraging prior knowledge about the model of the environment. By incorporating this prior, we can significantly improve sample efficiency, which enables us to conduct experiments on real robots.

Combining Model-Based and Model-Free Updates for Trajectory-Centric Reinforcement Learning
Arxiv, 2017
Y. Chebotar*, K. Hausman*, M. Zhang*, G. Sukhatme, S. Schaal, S. Levine
Generalizing Regrasping with Supervised Policy Learning
International Symposium on Experimental Robotics (ISER), 2016          
Y. Chebotar*, K. Hausman*, O. Kroemer, G. Sukhatme, S. Schaal
Self-Supervised Regrasping using Spatio-Temporal Tactile Features and Reinforcement Learning
International Conference on Intelligent Robots and Systems (IROS), 2016
Y. Chebotar, K. Hausman, Z. Su, G. Sukhatme, S. Schaal

Workshop Publications

Regrasping using Tactile Perception and Supervised Policy Learning
AAAI Symposium on Interactive Multi-Sensory Object Perception for Embodied Agents, 2017
Y. Chebotar, K. Hausman, Z. Su, G. Sukhatme, S. Schaal bibtex pdf

Supervised Policy Fusion with Application to Regrasping
IROS Workshop on Closed-loop Grasping and Manipulation: Challenges and Progress, 2016
Y. Chebotar*, K. Hausman*, O. Kroemer, G. Sukhatme, S. Schaal bibtex pdf




Interactive Perception

Recent approaches in Robotics are subsumed by the term Interactive Perception (IP). Within these approaches any kind of forceful interactions with the environment are used to simplify and enhance perception, thereby enabling robust perceptually-guided manipulation behaviors. IP has two benefits. First, physical interaction creates a novel sensory signal that would otherwise not be present. Second, by exploiting knowledge of the regularity in the combined space of sensory data and action parameters, the prediction and interpretation of this novel signal becomes simpler and more robust. For more details, see our survey paper.

Interactive Perception: Leveraging Action in Perception and Perception in Action
Submitted to IEEE Transactions on Robotics (T-RO), 2016
J. Bohg*, K. Hausman*, B. Sankaran*, O. Brock, D. Kragic, S. Schaal, G. Sukhatme
Active Articulation Model Estimation through Interactive Perception
International Conference on Robotics and Automation (ICRA), 2015
K. Hausman, S. Niekum, S. Osentoski , G. Sukhatme
Force Estimation and Slip Detection for Grip Control using a Biomimetic Tactile Sensor
International Conference on Humanoid Robotics (Humanoids), 2015
Z. Su, K. Hausman, Y. Chebotar, A. Molchanov, G. Loeb, G. Sukhatme, S. Schaal
Interactive Segmentation of Textured and Textureless Objects
Chapter in Handling Uncertainty and Networked Structure in Robot Control, L. Busoniu and L. Tamas (eds.), Springer, 2015
K. Hausman, D. Pangercic, Z. Marton, F. Belent-Benczedi, C. Bersch, M. Gupta, G. Sukhatme, M. Beetz
Tracking-based Interactive Segmentation of Textureless Objects
International Conference on Robotics and Automation (ICRA), 2013
Best Service Robotics Paper Finalist
K. Hausman, F. Balint-Benczedi, D. Pangercic, Z. Marton,
R. Ueda, K. Okada, M. Beetz

Workshop Publications

BiGS: BioTac Grasp Stability Dataset
ICRA Workshop on Grasping and Manipulation Datasets, 2016
Y. Chebotar, K. Hausman, Z. Su, A. Molchanov, O. Kroemer, G. Sukhatme, S. Schaal
website bibtex pdf

Slip Classification Using Tangential and Torsional Skin Distortions
on a Biomimetic Tactile Sensor

BMVA Workshop on Visual, Tactile and Force Sensing for Robot Manipulation, 2015
Z. Su, K. Hausman, Y. Chebotar, A. Molchanov, G. Loeb, G. Sukhatme, S. Schaal bibtex pdf

Slip Detection and Classification for Grip Control using Multiple Sensory Modalities
on a Biomimetic Tactile Sensor

IROS Workshop on Multimodal Sensor-Based Robot Control for HRI and Soft Manipulation, 2015
Z. Su, K. Hausman, Y. Chebotar, A. Molchanov, G. Loeb, G. Sukhatme, S. Schaal bibtex pdf

Towards Interactive Object Recognition
IROS 3rd Workshop on Robots in Clutter: Perception and Interaction in Clutter, 2014
K. Hausman, C. Corcos, J. Mueller, F. Sha, G. Sukhatme bibtex pdf

Segmentation of Cluttered Scenes through Interactive Perception
ICRA Workshop on Semantic Perception and Mapping for Knowledge-enabled Service Robotics, 2012
K. Hausman, C. Bersch, D. Pangercic, S. Osentoski, Z. Marton, M. Beetz bibtex pdf

Segmentation of Textured and Textureless Objects through Interactive Perception
RSS Workshop on Robots in Clutter: Manipulation, Perception and Navigation in Human Environments, 2012
C. Bersch, D. Pangercic, S. Osentoski, K. Hausman, Z. Marton, R. Ueda, K. Okada, M. Beetz bibtex pdf




Active Perception

In robotics, the research field of Active Perception pioneered the insight that perception is active and exploratory. In my research, I try to show that, state estimation (perception) can be significantly improved when considered jointly with control (action). I demonstrate my research results on various flying vehicles such as quadrotors.

Observability-Aware Trajectory Optimization for Self-Calibration with Application to UAVs
Robotics and Automation Letter (RA-L), 2017
K. Hausman, J. Preiss, G. Sukhatme, S. Weiss
Occlusion-Aware Multi-Robot 3D Tracking
International Conference on Intelligent Robots and Systems (IROS), 2016
K. Hausman, G. Kahn, S. Patil, J. Mueller, K. Goldberg, P. Abbeel, G. Sukhatme
Cooperative Multi-Robot Control for Target Tracking with Onboard Sensing
International Journal of Robotics Research (IJRR), 2015
K. Hausman, J. Mueller, A. Hariharan, N. Ayanian, G. Sukhatme
Self-Calibrating Multi-Sensor Fusion with Probabilistic Measurement Validation for Seamless Sensor Switching on a UAV
International Conference on Robotics and Automation (ICRA), 2016
K. Hausman, S. Weiss, R. Brockers, L. Matthies, G. Sukhatme
Cooperative Control for Target Tracking with Onboard Sensing
International Symposium on Experimental Robotics (ISER), 2014
K. Hausman, J. Mueller, A. Hariharan, N.Ayanian, G. Sukhatme

Workshop Publications

Observability-Aware Trajectory Optimization for Self-Calibration with Application to UAVs
RSS Workshop on Robot-Environment Interaction for Perception and Manipulation, 2016
K. Hausman, J. Preiss, G. Sukhatme, S. Weiss bibtex pdf

Optimization-based Cooperative Multi-Robot Target Tracking
with Reasoning about Occlusions

IROS Workshop on On-line Decision-Making in Multi-Robot Coordination, 2015
K. Hausman, G. Kahn, S. Patil, J. Mueller, K. Goldberg, P. Abbeel, G. Sukhatme bibtex pdf

Cooperative Multi-Robot Control for Target Tracking
with Efficient Switching of Onboard Sensing Topologies

IROS Workshop on Taxonomies of Interconnected Systems:
Topology in Distributed Robotics, 2014
K. Hausman, J. Mueller, A. Hariharan, N. Ayanian, G. Sukhatme bibtex pdf