Inverting Learned Dynamics Models for Aggressive Multirotor Control

Spitzer, Alexander; Michael, Nathan

doi:10.15607/rss.2019.xv.065

Cited by 5 publications

(6 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3) A learned acceleration model correction that is used in the feedback linearization controller to fully compensate for the unmodeled dynamics and external disturbances. This is an extension of the work in [5] for the feedback linearization controller. We experimentally analyze the effectiveness of the acceleration error compensation and input delay mitigation model in the feedback linearization controller on a real hardware quadrotor.…”

Section: Introductionmentioning

confidence: 82%

“…Yes ¸ildirek and Lewis [17] uses neural networks to learn the forward dynamics model, after which it is used in feedback linearization and [18] learns a forward model using a Gaussian Process (GP), whose uncertainty estimates are used to prove a convergence guarantee. Spitzer and Michael [5] learns a forward model and uses its derivatives to compensate for the disturbance dynamics. On the other hand, [19] and [20] learn an inverse model for feedback linearization.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Feedback Linearization for Quadrotors with a Learned Acceleration Error Model

Spitzer,

Michael

2021

Preprint

Self Cite

View full text Add to dashboard Cite

This paper enhances the feedback linearization controller for multirotors with a learned acceleration error model and a thrust input delay mitigation model. Feedback linearization controllers are theoretically appealing but their performance suffers on real systems, where the true system does not match the known system model. We take a step in reducing these robustness issues by learning an acceleration error model, applying this model in the position controller, and further propagating it forward to the attitude controller. We show how this approach improves performance over the standard feedback linearization controller in the presence of unmodeled dynamics and repeatable external disturbances in both simulation and hardware experiments. We also show that our thrust control input delay model improves the step response on hardware systems.

show abstract

Section: Introductionmentioning

confidence: 82%

Section: Related Workmentioning

confidence: 99%

Feedback Linearization for Quadrotors with a Learned Acceleration Error Model

Spitzer,

Michael

2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…A drawback of this method is being computationally demanding such that a ground computer is required to perform MPC calculations. Another work by [18] leveraged a learning based technique to produce adequate feedforward terms to perform accurate trajectory tracking in the presence of modeling and disturbance uncertainties. In a similar way to the other feedforward tuning methodologies, a regression algorithm learned a drag model based on flight data.…”

Section: B Related Workmentioning

confidence: 99%

“…In a similar way to the other feedforward tuning methodologies, a regression algorithm learned a drag model based on flight data. All these methods [13], [15], [17], [18] require extensive data collection and offline optimization to build up a drag or disturbance model, which limits the suitability for real-time adaptation. Also the optimized model can be biased towards the training data, and might under perform for unseen scenarios.…”

Section: B Related Workmentioning

confidence: 99%

Systematic Online Tuning of Multirotor UAVs for Accurate Trajectory Tracking Under Wind Disturbances and In-Flight Dynamics Changes

et al. 2022

View full text Add to dashboard Cite

The demand for accurate and fast trajectory tracking for multirotor Unmanned Aerial Vehicles (UAVs) have grown recently due to advances in UAV avionics technology and application domains. In many applications, the multirotor UAV is required to accurately perform aggressive maneuvers in challenging scenarios like the presence of external wind disturbances or in-flight payload changes. In this paper, we propose a systematic controller tuning approach based on identification results obtained by a recently developed Deep Neural Networks with the Modified Relay Feedback Test (DNN-MRFT) algorithm. We formulate a linear equivalent representation suitable for DNN-MRFT using feedback linearization. This representation enables the analytical investigation of different controller structures and tuning settings, and captures the non-linearity trends of the system. With this approach, the trade-off between performance and robustness in design was made possible which is convenient for the design of controllers of UAVs operating in uncertain environments. We demonstrate that our approach is adaptive and robust through a set of experiments, where accurate trajectory tracking is maintained despite significant changes to the UAV aerodynamic characteristics and the application of wind disturbance. Due to the model-based system design, it was possible to obtain low discrepancy between simulation and experimental results which is beneficial for potential use of the proposed approach for real-time model-based planning and fault detection tasks. We obtained RMSE of 3.59 cm when tracking aggressive trajectories in the presence of strong wind, which is on par with state-of-the-art.INDEX TERMS Unmanned aerial vehicles, system identification, adaptive and robust control, trajectory tracking, PID control, machine learning.

show abstract

“…Another approach is given a desired state, the control inputs are solved in an equation that includes the forward model. In [26], in order to improve the tracking performance of a quadrocopter, an acceleration error model is learned using the linear regression. It is then embedded into the acceleration model equation, in which the body angular acceleration command is solved for in real-time as the control inputs.…”

Section: ) Inverse Model Approachmentioning

confidence: 99%

Learning Dynamics for Improving Control of Overactuated Flying Systems

Zhang

Brunner

Ott

et al. 2020

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

Overactuated omnidirectional flying vehicles are capable of generating force and torque in any direction, which is important for applications such as contact-based industrial inspection. This comes at the price of an increase in model complexity. These vehicles usually have non-negligible, repetitive dynamics that are hard to model, such as the aerodynamic interference between the propellers. This makes it difficult for high-performance trajectory tracking using a model-based controller. This paper presents an approach that combines a datadriven and a first-principle model for the system actuation and uses it to improve the controller. In a first step, the first-principle model errors are learned offline using a Gaussian Process (GP) regressor. At runtime, the first-principle model and the GP regressor are used jointly to obtain control commands. This is formulated as an optimization problem, which avoids ambiguous solutions present in a standard inverse model in overactuated systems, by only using forward models. The approach is validated using a tilt-arm overactuated omnidirectional flying vehicle performing attitude trajectory tracking. The results show that with our proposed method, the attitude trajectory error is reduced by 32% on average as compared to a nominal PID controller.

show abstract

Inverting Learned Dynamics Models for Aggressive Multirotor Control

Cited by 5 publications

References 39 publications

Feedback Linearization for Quadrotors with a Learned Acceleration Error Model

Feedback Linearization for Quadrotors with a Learned Acceleration Error Model

Systematic Online Tuning of Multirotor UAVs for Accurate Trajectory Tracking Under Wind Disturbances and In-Flight Dynamics Changes

Learning Dynamics for Improving Control of Overactuated Flying Systems

Contact Info

Product

Resources

About