Dr. Mikko Lauri

My research investigates active sensing: the ability of autonomous agents, such as robots, to control
how they use their sensors to help them gain more information about their environment. Solving
active sensing problems answers questions such as where to look next, which sensors to activate
and when, and so on. This leads to more reliable estimation of the robot’s own state and the state of
the environment, and thereby to safer and more efficient autonomous systems.
I apply decision and control theory, Bayesian statistics, and information theory to active sensing
problems. Application areas include active perception for object discovery, environmental
monitoring and exploration, visual attention, and active simultaneous localization and mapping
(SLAM), among others.
I joined the Computer Vision group in August 2016. Prior to that, I completed my PhD at Tampere
University of Technology, in Tampere, Finland.

[Google Scholar profile]

2018

• Ge Gao, Mikko Lauri, Jianwei Zhang, Simone Frintrop: Occlusion Resistant Object Rotation Regression from Point Cloud Segments, Proceeding of the ECCV workshop on Recovering 6D Object Pose, 2018, [PDF], [arXiv]

2017

• Ge Gao, Mikko Lauri, Jianwei Zhang, Simone Frintrop: Saliency-guided Adaptive Seeding for Supervoxel Segmentation,  IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2017, [PDF], [arXiv], [code and more details]
  
• Mikko Lauri, Eero Heinänen, Simone Frintrop: Multi-Robot Active Information Gathering with
  Periodic Communication, IEEE International Conference on Robotics and Automation
  (ICRA), 2017

Image: ICRA17_policy.png

Description: We propose decentralized partially observable Markov decision process (Dec-
POMDP) with an information-theoretic reward function. We apply it to a information gathering task where a team of robots collaboratively maximizes the amount of information about the state of the system.

• Mikko Lauri, Simone Frintrop: Object proposal generation applying the distance dependent Chinese
  restaurant process, Scandinavian Conference on Image Analysis (SCIA), 2017, PDF, Software: https://github.com/laurimi/ddcrp-gibbs, Preprint on arXiv: https://arxiv.org/abs/1704.03706

Image: SCIA17_process.png

Description: We develop a system for generating object proposals, regions of an image that likely
correspond to objects. Through use of non-parametric Bayesian inference, we estimate how likely
an proposal is according to properties such as internal similarity and contrast to surroundings. We rank the resulting proposals based on a score derived from the Gestalt principles that model human object perception.

2016

• Mikko Lauri, Risto Ritala: Planning for robotic exploration based on forward simulation, Robotics
  and Autonomous Systems, Volume 83, September 2016, Pages 15-31,
  http://dx.doi.org/10.1016/j.robot.2016.06.008 .
  Preprint: https://arxiv.org/pdf/1502.02474.pdf
  Video: https://www.youtube.com/watch?v=-7c_RaZ-KTU
  Software (github): https://github.com/laurimi/ase_exploration

Image: RAS16_exploration.png

Description: We propose an information-theoretic planning approach to autonomous exploration of an environment by a robot. We show that in specific cases, applying an information-theoretic objective function for choosing the next robot trajectories can outperform traditional approaches such as frontier exploration.

2015

• Mikko Lauri, Risto Ritala: Optimal sensing via multi-armed bandit relaxations in mixed
  observability domains, IEEE International Conference on Robotics and Automation (ICRA), 2015,
  pp. 4807-4812
  Preprint: https://arxiv.org/pdf/1603.04586.pdf 

Image: ICRA15_environment.png

Description: A robot is moving in a graph-like environment, monitoring the state of a set of random variables at each of the graph nodes. We describe a connection between this problem and a simpler multi-armed bandit (MAB) problem. The solution of the MAB problem is applied to guide a search algorithm for solving the original information gathering problem, reducing the required computational effort.