Home >Projects >Deep Grid Net

ROVIS Robotics Summer School

ByROVIS team

Organisation

Target group: Students, PhD students, Professors
Lecturers: Claudiu Pozna, Sorin Grigorescu, Liviu Marina, Tiberiu Cocias, Gigel Macesanu
Language: English

Course aim

The main objective of the course is to give a general understanding of main concepts related to robot modeling and control, deep learning, robot vision and robotic operating systems. During this summer school, all students will start with the basics of manipulation models and mobile robots. Then, an introduction on robots operating system will be presented, followed by general topics about Machine Learning and Computer Vision theory. At the end of the course an overview of available algorithms for objects detection and recognition will be presented.

Content

Lecture Module Introduction Advanced
1 Modeling and Control of Robotic Systems (Claudiu Pozna) Manipulators Models (8 hours):
  • Mathematical descriptions (models) of manipulators.
  • Starting with a Mathematical refresher we increase step by step the accuracy of knowledge.
  • Manipulator models: geometric, kinematic and dynamic model.
  • Programing and simulations (in Matlab).
Mobile Robots Models (10 hours):
  • Mathematical description of an autonomous car.
  • Trajectory definition; Comparator; PID controller; Car model; Sensors and filters.
  • Programing and simulations (in Matlab).
 Control of manipulators (8 hours):
  • Control Design refresher: control methods.
  • DC motor control; Control based on Jacobian; Dynamic control.
  • Programing and simulations (in Matlab).
2 Robot Operating System (ROS) (Gigel Macesanu) Introduction in ROS (8 hours):
  • ROS Architecture.
  • ROS Abstraction.
  • ROS Node and Topic.
  • ROS Communication.
  • ROS Simulations - Basig shapes in Rviz.
 Advanced Simulation in ROS (10 hours):
  • Introduction in Gazebo.
  • Controlling the Pioneer 3-DX robot.
3 Machine Learning in Robotics (Sorin Grigorescu) Linear Regression (4 hours):
  • Model representation
  • Cost function
  • Gradient descenet
Logistic Regression (4 hours):
  • Hypothesis representation
  • Cost function
  • Gradient descenet
 Neural Networks (4 hours):
  • Network representation and feedforward processing
  • Backpropagation method for training
Convolutional Neural Networks (4 hours):
  • Architecture
  • Convolution
  • Pooling
4 Robot Vision (Cocias Tiberiu) Introduction to computer vision (4 hours):
  • Image formation and filtering
  • Image representation and noise
  • Spatial filtering
Object recognition (4 hours):
  • Template matching
  • Region segmentation
  • Edge detection
 Optics and 3D Reconstruction (4 hours):
  • Ideal camera model
  • Camera calibration
  • Stereo Vision
  • Correspondence matching
  • Point clouds processing
Object tracking (4 hours):
  • Optical flow
  • Dynamic models for object tracking