Bayesian Estimation for Flight Robotics (Lecture + Practical)

Format: 24 Lectures (12 weeks) + 24 Practical Sessions

Occurrence: Offered every summer semester 

Course Level: Master Students (Typically in their 1st or 2nd semester)



State estimation in autonomous UAVs, or flying robots, is a critical functionality. It allows a UAV to accurately localize itself and estimate the positions of a tracked target (from the UAV). Without robust state estimation, any of the other functionalities like planning, decision making and even low-level control can be jeopardized. The goal of this course is to provide students with a deep understanding of Bayesian estimation techniques and how they can be applied for UAV state estimation and target tracking. Students should be able to appreciate parametric ad non-parametric forms of estimation and their pros and cons. Finally, they should be able to implement various Bayesian filters for online applications, both in simulation and in a real UAV.


Lecture Contents

  • Bayesian filtering: Introduction to recursive state estimation, concept of Markov assumption and basic Bayes filter.
  • Sensor models:
        * Probabilistic sensor models: e.g., laser range finders, RGB cameras.
        * Modeling raw sensor measurements and feature-based approaches.
        * Overview of camera projection models e.g., perspective, equidistant, etc.
        * Barometric pressure based altitude sensors.
        * Differential pressure airspeed sensors     
  • Motion and Sensor models:
        * Probabilistic approaches to robot motion modeling
        * Velocity motion model
        *  Fixed wing UAV with dead reckoning
  • Parametric filters: (Extended) Kalman filter and Information filter
        * Basics of KF, EKF, derivation, assumptions and properties.
        * Canonical representations, their advantages, Information filter (IF) and extended versions. Overview of related state-of-the-art techniques.
  • Non-Parametric filters: Histogram filter and Particle filter and variants
        * Why non-parametric filters? Advantages (multiple modalities, nonlinearities) and Disadvantages (curse of dimensionality). Discretization and Histogram filtering.
    * Particle filters (PFs): concepts and properties. Discussion of state-of-the-art PF-based methods.

Practical Contents


  • [Topic 1] Introduction to to ROS, Gazebo and Python
  • [Topic 2] Intro to ‘how to use our UAV environment’ and code a node — Sensors.
  • [Topic 3] State estimation and control -- Code ROS nodes for UAV localization using on-board IMU and simulated GPS
  • [Topic 4] Visual Tracking -- Code ROS node to track the state of a moving target from a UAV.