Mario Luca Fravolini

Francesco Crocetti, G. Costante, M. L. Fravolini et al.

The accurate online estimation of the road-friction coefficient is an essential feature for any advanced brake control system. In this study, a data-driven scheme based on a MLP Neural Net is proposed to estimate the optimum friction coefficient as a function of windowed slip-friction measurements. A stochastic NN weights drop-out mechanism is used to online estimate the confidence interval of the estimated best friction coefficient thus providing a characterization of the epistemic uncertainty associated to the NN block. Open loop and closed loop simulations of the landing phase of an aircraft on an unknown surface are used to show the potentiality and efficacy of the proposed robust friction estimation approach.

F. Crocetti, G. Costante, M. L. Fravolini et al.

The performances of braking control systems for robotic platforms, e.g., assisted and autonomous vehicles, airplanes and drones, are deeply influenced by the road-tire friction experienced during the maneuver. Therefore, the availability of accurate estimation algorithms is of major importance in the development of advanced control schemes. The focus of this paper is on the estimation problem. In particular, a novel estimation algorithm is proposed, based on a multi-layer neural network. The training is based on a synthetic data set, derived from a widely used friction model. The open loop performances of the proposed algorithm are evaluated in a number of simulated scenarios. Moreover, different control schemes are used to test the closed loop scenario, where the estimated optimal slip is used as the set-point. The experimental results and the comparison with a model based baseline show that the proposed approach can provide an effective best slip estimation.

Silvia Cascianelli, Gabriele Costante, Alessandro Devo et al.

Natural Language Video Description (NLVD) has recently received strong interest in the Computer Vision, Natural Language Processing (NLP), Multimedia, and Autonomous Robotics communities. The State-of-the-Art (SotA) approaches obtained remarkable results when tested on the benchmark datasets. However, those approaches poorly generalize to new datasets. In addition, none of the existing works focus on the processing of the input to the NLVD systems, which is both visual and textual. In this work, it is presented an extensive study dealing with the role of the visual input, evaluated with respect to the overall NLP performance. This is achieved performing data augmentation of the visual component, applying common transformations to model camera distortions, noise, lighting, and camera positioning, that are typical in real-world operative scenarios. A t-SNE based analysis is proposed to evaluate the effects of the considered transformations on the overall visual data distribution. For this study, it is considered the English subset of Microsoft Research Video Description (MSVD) dataset, which is used commonly for NLVD. It was observed that this dataset contains a relevant amount of syntactic and semantic errors. These errors have been amended manually, and the new version of the dataset (called MSVD-v2) is used in the experimentation. The MSVD-v2 dataset is released to help to gain insight into the NLVD problem.