End-to-End Neural Network for Vehicle Dynamics Modeling

Hermansdorfer, Leonhard; Trauth, Rainer; Betz, Johannes; Lienkamp, Markus

doi:10.1109/cist49399.2021.9357196

Cited by 18 publications

(6 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…MLP [ 24 , 25 , 57 ]: The multilayer perceptron (MLP) is a commonly used approach for building data-driven vehicle dynamics models. The structure of the MLP used in this study was inspired by [ 24 ] and was chosen as {22, 32, 32

5, 64, 128, 128, 64, 32, 6}.…”

Section: Resultsmentioning

confidence: 99%

“…LSTM [ 24 , 57 , 58 ]: Long Short-Term Memory (LSTM) is another classic method used to build data-driven dynamics models, which is employed for predicting the acceleration and angular velocity of vehicles [ 57 , 58 ]. LSTM is known for its ability to capture temporal dependencies in the data.…”

Section: Resultsmentioning

confidence: 99%

“…However, there was a difference in the loss calculation method. Different from the loss calculation method established by PINN, MLP and LSTM used the IMU measurements to build data-driven vehicle models [ 25 , 57 , 58 ]. These methods commonly assumed that the ground truth states of the vehicle were accessible.…”

Section: Resultsmentioning

confidence: 99%

See 2 more Smart Citations

Vehicle State Estimation Combining Physics-Informed Neural Network and Unscented Kalman Filtering on Manifolds

Tan

Cai

Wang

et al. 2023

Sensors

View full text Add to dashboard Cite

This paper proposes a novel vehicle state estimation (VSE) method that combines a physics-informed neural network (PINN) and an unscented Kalman filter on manifolds (UKF-M). This VSE aimed to achieve inertial measurement unit (IMU) calibration and provide comprehensive information on the vehicle’s dynamic state. The proposed method leverages a PINN to eliminate IMU drift by constraining the loss function with ordinary differential equations (ODEs). Then, the UKF-M is used to estimate the 3D attitude, velocity, and position of the vehicle more accurately using a six-degrees-of-freedom vehicle model. Experimental results demonstrate that the proposed PINN method can learn from multiple sensors and reduce the impact of sensor biases by constraining the ODEs without affecting the sensor characteristics. Compared to the UKF-M algorithm alone, our VSE can better estimate vehicle states. The proposed method has the potential to automatically reduce the impact of sensor drift during vehicle operation, making it more suitable for real-world applications.

show abstract

5, 64, 128, 128, 64, 32, 6}.…”

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Vehicle State Estimation Combining Physics-Informed Neural Network and Unscented Kalman Filtering on Manifolds

Tan

Cai

Wang

et al. 2023

Sensors

View full text Add to dashboard Cite

show abstract

“…In their work, one hidden layer was applied with 150 GRU (Gated Recurrent Unit) neurons. However, there is no future prediction available for the outputs mentioned [ 23 ].…”

Section: Literature Reviewmentioning

confidence: 99%

Prediction for Future Yaw Rate Values of Vehicles Using Long Short-Term Memory Network

Kontos

Kránicz

Vathy-Fogarassy

2023

Sensors

View full text Add to dashboard Cite

Currently, electric mobility and autonomous vehicles are of top priority from safety, environmental and economic points of view. In the automotive industry, monitoring and processing accurate and plausible sensor signals is a crucial safety-critical task. The vehicle’s yaw rate is one of the most important state descriptors of vehicle dynamics, and its prediction can significantly contribute to choosing the correct intervention strategy. In this article, a Long Short-Term Memory network-based neural network model is proposed for predicting the future values of the yaw rate. The training, validating and testing of the neural network was conducted based on experimental data gathered from three different driving scenarios. The proposed model can predict the yaw rate value in 0.2 s in the future with high accuracy, using sensor signals of the vehicle from the last 0.3 s in the past. The R2 values of the proposed network range between 0.8938 and 0.9719 in the different scenarios, and in a mixed driving scenario, it is 0.9624.

show abstract

“…In [5], they proposed a simple neural network to replace a single-track vehicle model and used it to generate feedforward control signals. Similarly, in [6], they designed Deep Neural Networks (DNN) as a model approximator to identify the vehicle dynamics model in an end-to-end learning fashion. However, while DNN is an efficient way to approximate nonlinear systems, it is difficult to integrate with non-learning model-based methods, which are reliable in real-world applications.…”

Section: Introductionmentioning

confidence: 99%

Data-Driven Model Identification via Hyperparameter Optimization for Autonomous Racing Systems

Hyunki¹,

Chung²,

Shim³

2023

Preprint

View full text Add to dashboard Cite

In this letter, we propose a model identification method via hyperparameter optimization (MIHO). Our method adopts an efficient explore-exploit strategy to identify the parameters of dynamic models in a data-driven optimization manner. We utilize MIHO for model parameter identification of the AV-21, a full-scaled autonomous race vehicle. We then incorporate the optimized parameters for the design of model-based planning and control systems of our platform. In experiments, the learned parametric models demonstrate good fitness to given datasets and show generalization ability in unseen dynamic scenarios. We further conduct extensive field tests to validate our model-based system. The tests show that our race systems leverage the learned model dynamics and successfully perform obstacle avoidance and high-speed driving over 200km/h at the Indianapolis Motor Speedway and Las Vegas Motor Speedway. The source code for MIHO and videos of the tests are available at https://github.com/hynkis/MIHO.

show abstract

End-to-End Neural Network for Vehicle Dynamics Modeling

Cited by 18 publications

References 12 publications

Vehicle State Estimation Combining Physics-Informed Neural Network and Unscented Kalman Filtering on Manifolds

Vehicle State Estimation Combining Physics-Informed Neural Network and Unscented Kalman Filtering on Manifolds

Prediction for Future Yaw Rate Values of Vehicles Using Long Short-Term Memory Network

Data-Driven Model Identification via Hyperparameter Optimization for Autonomous Racing Systems

Contact Info

Product

Resources

About