Robust and Sample-Efficient Estimation of Vehicle Lateral Velocity Using Neural Networks With Explainable Structure Informed by Kinematic Principles

Abstract

This paper presents kinematics-structured neural networks (KS-NN) for the lateral speed estimation of vehicles. The internal structure of the networks is designed to incorporate the kinematic principles, enhancing the physical explainability and generalization capacity. Both the internal structure and training method are devised for better generalization performance. Various linear and nonlinear variants of our estimator are assessed for accuracy and robustness. The approach is validated using an openly accessible dataset with two race cars. The performance of the novel networks is evaluated against Luenberger, neural network, factor graph and Kalman filter estimators from the literature. In comparison with a Luenberger kinematic observer, our networks improve noise rejection, and overcome the well-known observability problem for low yaw rates. Compared to existing neural network estimators, our networks exhibit better generalization capacity, are more sampleefficient, require fewer learnable parameters, and their structure is physically explainable.

Multi-Layered Local Dynamic Map for a Connected and Automated In-Vehicle System

“…Given that the incoming data of the sensor system was already filtered, and we were working in post-processing, we decided to take it "as is", as this would be the case for incoming measurements received via CP. However, once we deploy our solution on the prototype vehicle and use it online, we will consider the use of additional advanced estimation techniques, such as the robust lateral velocity estimator developed by Mauro Da Lio et al in [39], to reduce the uncertainty of the GNSS system.…”

“…Although the prediction window is not wide, 0.4 s is enough to fill in the blanks that may emerge from communication delays/interruptions or temporary occlusions of entities, just to name a few. Improving the positioning of the ego vehicle and/or of detected entities, for example, using additional advanced estimation techniques like the one mentioned a few lines above [39], will undoubtedly improve the data persistence capabilities as well.…”

“…• Enhancing the matching algorithm by varying parameters based on the type of entity detected; • Converting the LDM to C++ and using the igraph package [40] to manage the graph database directly in the API; • Using the LDM to augment the capabilities of our motion planning framework for at-limit handling of racing vehicles [41]; • Using additional advanced estimation techniques [39] to increase the accuracy of the GNSS system; • Deploying the developed solution on the prototype Automated Vehicle from CRF Trento Branch and testing it online.…”

A Physics-Driven Artificial Agent for Online Time-Optimal Vehicle Motion Planning and Control

“…The structure of F N is inspired by the theory of neuro-fuzzy local models [48], [49], and we design the local models by leveraging the prior knowledge VOLUME 11, 2023 of the vehicle dynamic laws. We used a conceptually similar approach in [19], [50], and [51]. The learnable weights {A 1 , .…”

Fast Planning and Tracking of Complex Autonomous Parking Maneuvers With Optimal Control and Pseudo-Neural Networks

“…In comparison with other papers using neural networks for motion tracking of parking maneuvers [28], the specific structure devised for pNN requires smaller training datasets to produce accurate predictions. The structure of pNN is partly inspired by the neural models that we presented in [31], [42], [43], but the model of this paper is designed to capture the nonlinear steering dynamics at low speed.…”