aiqbench analysis
This research introduces an energy-based regularization technique to enhance neural model predictive control (MPC) for aerial robots, addressing the common issue where deep learning models fail to capture physical properties like inertia. By integrating this loss function into the training of residual dynamics, the method achieves significant improvements in flight stability and tracking accuracy compared to standard analytical or unregularized neural approaches. The work represents a meaningful step toward physically consistent AI control systems that better respect conservation laws.
Energy-based Regularization for Learning Residual Dynamics in Neural MPC for Omnidirectional Aerial Robots
arXiv:2604.14678v2 Announce Type: replace-cross Abstract: Data-driven Model Predictive Control (MPC) has lately been a core research subject in the field of control theory. The combination of an optimal control framework with deep learning paradigms opens up the possibility to accurately tracking control tasks without the need for complex analytical models. However, the system dynamics are often nuanced and the neural model lacks the potential to understand physical properties such as inertia and conservation of energy. In this work, we propose a novel energy-based regularization loss function which is applied to the training of a neural model that learns the residual dynamics of an omnidirectional aerial robot. Our energy-based regularization encourages the neural network to learn stabilizing control corrections. Without the regularization the network misses physical context. The learned residual dynamics are then integrated into the MPC framework. The positional mean absolute error (MAE) are shown to be improved in three real-world experiments by 23% compared to the analytical MPC. We also compare our method to a standard neural MPC implementation without regularization and primarily achieve an increased flight stability implicitly due to smoothening out the acceleration of the robot and thus up to 15% lower MAE. A complimentary video is available at: https://youtu.be/E3XG979NXHk. We share our code under: https://github.com/johanneskbl/jsk_aerial_robot/tree/develop/neural_MPC.
Abstract:Data-driven Model Predictive Control (MPC) has lately been a core research subject in the field of control theory. The combination of an optimal control framework with deep learning paradigms opens up the possibility to accurately tracking control tasks without the need for complex analytical models. However, the system dynamics are often nuanced and the neural model lacks the potential to understand physical properties such as inertia and conservation of energy. In this work, we propose a novel energy-based regularization loss function which is applied to the training of a neural model that learns the residual dynamics of an omnidirectional aerial robot. Our energy-based regularization encourages the neural network to learn stabilizing control corrections. Without the regularization the network misses physical context. The learned residual dynamics are then integrated into the MPC framework. The positional mean absolute error (MAE) are shown to be improved in three real-world experiments by 23% compared to the analytical MPC. We also compare our method to a standard neural MPC implementation without regularization and primarily achieve an increased flight stability implicitly due to smoothening out the acceleration of the robot and thus up to 15% lower MAE. A complimentary video is available at: this https URL. We share our code under: this https URL.
Systems and Control (eess.SY); Robotics (cs.RO)
arXiv:2604.14678 [eess.SY]
(or arXiv:2604.14678v2 [eess.SY] for this version)
https://doi.org/10.48550/arXiv.2604.14678 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 303 words
Related coverage
Dynamical and Optimization Trade-offs of Levi--Civita Coordinates for Learned Close-Encounter Dynamics
arXiv · Jul 23, 2026
COFM: Consistent Optimal Transport Flow Matching via Partially Input Convex Neural Networks
arXiv · Aug 28, 2026
A Physics-Inspired Optimizer: Velocity Regularized Adam
arXiv · Jun 12, 2026
In-Context Residual Calibration for Uncertainty Quantification of Energy Time Series over Graphs
arXiv · Oct 8, 2026