State Estimation for Legged Robots


Author / Producer

Date

2021-07

Publication Type

Bachelor Thesis

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

In the advent of the omnipresence of robotics in our day to day lives, progressions are made in this field continuously. In particular, the recent developments achieved within the area of legged robotics substantiate their growing importance for present and future applications. Robust and accurate state estimation capabilities are essential to the feedback control that enables legged robots to function within their expanding range of deployment. This performance is achieved by employing sensor fusion frameworks that combine multiple measurements to produce an estimate of the robots state. The approaches discussed in this thesis focus on using information provided by proprioceptive sensors to maintain robustness and fast estimation times. Further, the approaches employ sensor fusion by means of stochastic Kalman Filtering methods. The challenges presented by state estimation for legged robots however lead to estimation approaches that entail ever increas- ing implementation complexity to guarantee the accuracy of the estimate. Further, the amount of research being done within this field steadily increases the different state estimation techniques available for legged robots. While substantial progress is being made, not every increase in com- plexity is justified by its increase in accuracy. The significance of the robot’s estimation capability emphasizes the need for constant review and comparison between multiple state-of-the-art estimation approaches. This is further asserted by the multitude of techniques proposed, as well as by the tendency to higher implementation com- plexity. Building on the analysis of previous work, this thesis aims at implementing, combining and compar- ing different state-of-the-art state estimation approaches for legged robots. The implementations are tested in simulation and the results are evaluated. An emphasis is placed on the accuracy with which the approaches handle the challenges of state estimation for legged robots, as well as the level of implementation complexity they require to do so. As a result of this thesis, a Two Stage Estimation framework that separates linear and nonlinear estimation is proposed, which is based on prior work. The employment of an Unscented Kalman Filter for orientation estimation and a Classical Kalman Filter for position and velocity estimation is suggested, outlining the path to simpler and more accurate state estimation approaches for legged robots.

Publication status

published

External links

Editor

Contributors

Examiner : Kang, Dongho
Examiner : Coros, Stelian

Book title

Journal / series

Volume

Pages / Article No.

Publisher

ETH Zurich

Event

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

legged robotics; state estimation; robotics; Quadruped Locomotion

Organisational unit

09620 - Coros, Stelian / Coros, Stelian check_circle

Notes

Funding

Related publications and datasets