Curriculum Learning for Vehicle Lateral Stability Estimations

Abstract

Precise estimations of the roll and sideslip angles of autonomous vehicles are essential for autonomous driving, which requires further information about the vehicle state. As such, novel deep learning approaches have been introduced for this purpose. However, the majority of deep learning works focusing on vehicle dynamics estimations have yet to delve into learning strategies specifically for this task. Here, we argue that simply applying an adequate learning strategy to the task can boost the estimation performance. In this paper, we propose a simple yet effective curriculum learning strategy for better estimations of the roll and sideslip angles simultaneously. In addition, we compare our curriculum using a self-taught scoring function with a curriculum sorted by prior human knowledge, demonstrating its superiority. The proposed method outperforms the non-curriculum method by a large margin (up to a 16.5% decrease for sideslip as validation and 3.7% on a test), especially with regard to cornering (up to a 4% decrease). INDEX TERMSCurriculum learning, deep learning based estimator, roll angle, sensor fusion, sideslip angle, vehicle pose estimation.

Influence of Snowy and Icy Weather on Vehicle Sideslip and Rollover: A Simulation Approach

“…A lateral offset distance of more than 30 cm is usually suggestive of a vehicle sideslip. Roll angle, sideslip angle, side friction coefficient, and yaw angle are additional important indicators for assessing vehicle lateral stability [40][41][42][43].…”

Smooth Model Predictive Path Integral Control Without Smoothing

“…The Mean Squared Error (MSE) loss was used as the loss function and the Adam optimization was used for mini-batch gradient descent. We collected a human-controlled driving dataset in a manner similar to that utilized in our prior work on the training of a dynamics model [23]. The vehicle state is defined as x = [v x , v y , r] T , where v x and v y are the longitudinal and lateral velocities, and r is the yaw rate.…”

TOAST: Trajectory Optimization and Simultaneous Tracking Using Shared Neural Network Dynamics

“…We also evaluated the performances of different controllers in a high-fidelity vehicle simulator, the IPG CarMaker. It has been widely used to validate precisely nonlinear vehicle dynamics [21]- [23]. We built a race track with a length of 1016 m, two moderate curves, and four sharp curves (see Fig.…”

“…We collected a human-controlled driving dataset, with a data rate of 10 Hz, to train our network by expanding the methods used in our previous work [23]. We found that the dataset should comprise three types of distinct maneuvers in order for the neural network to accurately represent the vehicle dynamics for various friction conditions: i) Zig-zag driving at low speeds (20 − 25 km/h) on the race track, in both clockwise and counter-clockwise directions.…”