Pavel Trutman

I am a Ph.D. student advised by Tomas Pajdla at CIIRC, the intitute of the Czech Technical University in Prague. Currently, I work on application of techniques from semidefinite programming to polynomial optimization problems arising from computer vision and robotics.

Previously, I have received my master degree from Czech Technical University in Prague by defending the thesis Semidefinite Programming for Geometric Problems in Computer Vision. I have spent two summers in France. The first at LAAS-CNRS in Toulouse with Didier Henrion and the second in Willow team at INRIA in Paris with Josef Sivic.


Publications

Globally Optimal Solution to Inverse Kinematics of 7DOF Serial Manipulator
Pavel Trutman, Mohab Safey El Din, Didier Henrion, Tomas Pajdla
ArXiv preprint, 2020
Semidefinite Programming for Geometric Problems in Computer Vision
Pavel Trutman
Master's thesis, 2018
Minimal Problem Solver Generator
Pavel Trutman
Bachelor thesis, 2015

Talks

Globally Optimal Solution to Inverse Kinematics of 7DOF Serial Manipulator
Invited talk
Optimization Workshop, CIIRC, Prague, 21. 10. 2020
Globally Optimal Solution to Inverse Kinematics of 7DOF Serial Manipulator
Invited talk
IMPACT Seminar, CIIRC, Prague, 14. 1. 2020
Globally Optimal Solution to Inverse Kinematics of 7DOF Serial Manipulator
Invited talk
PGMO Days 2019, EDF Lab, Paris-Saclay, 3. – 4. 12. 2019
Ioannis Z. Emiris: A General Solver Based on Sparse Resultants
Reading group
Reading group in Pattern Recognition and Computer Vision, FEE, CTU in Prague, Prague, 30. 8. 2019

Software

Global 7DOF IKT solver

Solver for inverse kinematics tasks for general serial manipulators with 7 revolute joints. It combines symbolical and numerical approaches to find a global solution to a quadratic polynomial objective function. A toolbox for Matlab.

Polyopt

A Python package for modelling and solving moment LMI relaxations of polynomial optimization problems. The semidefinite solver is a stand-alone basic implementation of a primal interior-point method.

Automatic generator

Automatic generator of Gröbner basis solvers.

F4 Algorithm

An implementation of the F4 Algorithm in Maple.


Recent projects

IMPACT: Intelligent Machine Perception Project

The IMPACT project focuses on fundamental and applied research in computer vision, machine learning and robotics to develop machines that learn to perceive, reason, navigate and interact with complex dynamic environments.

LADIO: Live Action Data Input / Output

We build a fully integrated software for 3D reconstruction, photomodeling and camera tracking. We aim to provide a strong software basis with state-of-the-art computer vision algorithms that can be tested, analyzed and reused. Links between academia and industry is a requirement to provide cutting-edge algorithms with the robustness and the quality required all along the visual effects and shooting process.


Teaching assistant

Advanced Robotics
Winter 2018

We will explain some fundamental notions appearing in advanced robotics. We shall, e.g., learn how to solve the inverse kinematics task of a general serial manipulator with 6 degrees of freedom. There is a general solution to this problem but it can't easily be obtained by elementary methods. We shall present some more advanced algebraic tools for solving algebraic equations. We will also pay special attention to representing and parameterizing rotations and motions in 3D space. We will solve simulated problems as well as problems with real data in labs and assignments.

Geometry of Computer Vision and Graphics
Summer 2018

We will explain the basics of Euclidean, Affine and Projective geometry and show how to measure distances and angles in a scene from its images. We will introduce a model of the perspective camera, explain how images change when moving a camera and show how to find the camera pose from images. We will demonstrate the theory in practical tasks of panorama construction, finding the camera pose, adding a virtual object to a real scene and reconstructing a 3D model of a scene from its images.


Research stays

Mixed Reality & AI Zürich Lab, Microsoft, Zürich
September to November 2019, Johannes L. Schönberger

Main topic was to improve and generalize solvers for rolling shutter camera localization.

Willow team, INRIA, Paris
Summer 2016, Josef Sivic

Research in deep learning for algebraic geometry:

  • Investigation how deep neural networks could be used to train fast solvers (i.e. mappings of input coefficients to output solutions) for multi-view geometry problems.

  • Generating examples by simulations and use slow algebraic solvers to obtain data that would be efficiently "remembered and interpolated" by Neural Networks.

  • Validating the proposed method on synthetic and real data.

LAAS-CNRS, University of Toulouse, Toulouse
Summer 2015, Didier Henrion

Semidefinite programming for polynomial optimization in computer vision.


Contact