Premium
A framework for maximum likelihood parameter identification applied on MAVs
Author(s) -
Burri Michael,
Bloesch Michael,
Taylor Zachary,
Siegwart Roland,
Nieto Juan
Publication year - 2018
Publication title -
journal of field robotics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.152
H-Index - 96
eISSN - 1556-4967
pISSN - 1556-4959
DOI - 10.1002/rob.21729
Subject(s) - computer science , modular design , identification (biology) , key (lock) , inertial measurement unit , convergence (economics) , estimation theory , rotor (electric) , data mining , real time computing , control engineering , engineering , algorithm , artificial intelligence , mechanical engineering , botany , computer security , economics , biology , economic growth , operating system
With the growing availability of agile and powerful micro aerial vehicles (MAVs), accurate modeling is becoming more important. Especially for highly dynamic flights, model‐based estimation and control combined with a good simulation framework is key. While detailed models are available in the literature, measuring the model parameters can be a time‐consuming task and requires access to special equipment or facilities. In this paper, we propose a principled approach to accurately estimate physical parameters based on a maximum likelihood (ML) estimation scheme. Unlike many current methods, we make direct use of both raw inertial measurement unit measurements and the rotor speeds of the MAV. We also estimate the spatial‐temporal alignment to a modular pose sensor. The proposed ML‐based approach finds the parameters that best explain the sensor readings and also provides an estimate of their uncertainty. Although we derive the proposed method for use with an MAV, the approach is kept general and can be extended to other sensors or flying platforms. Extensive evaluation on simulated data and on real‐world experimental data demonstrates that the approach yields accurate estimates and exhibits a large region of convergence. Furthermore, we show that the estimation can be performed using only on‐board sensing, requiring no external infrastructure.