This paper studies neural network models of vehicle dynamics. We consider both models with a generic layer architecture and models with specialised topologies that hard-wire physics principles. Network pre-wiring is limited to universal laws; hence it does not limit the network modelling abilities on one side but allows more robust and interpretable models on the other side. Four different network types (with and without pre-wired structure, recursive and non-recursive) are compared for the longitudinal dynamics of a car with gears and two controls (brake and engine). Results show that pre-wiring effectively improves the performance. Non-recursive networks also look to be preferable for several reasons.

Non-exhaust emissions, generated by the wear of brake systems, tires, roads, clutches, and road resuspension, are responsible for a large part of airborne pollutants in urban areas. Brake wear accounts for 55% of non-exhaust emissions and significantly contributes to urban health diseases related to air pollution. A major part of the studies reported in the scientific literature are focused on experimental methods to sample and characterize brake wear particles in a reliable, representative, and repeatable way. In this framework, simulation is an important tool, which makes it possible to give interpretations of the experimental results, formulate new testing approaches, and predict the emission produced by brakes. The present comprehensive literature review aims to introduce the state of the art of the research on the different aspects of airborne wear debris resulting from brake systems which can be used as inputs in future simulation models. In this review, previous studies focusing on airborne emissions produced by brake systems are investigated in three main categories: the subsystem level, system level, and environmental level. As well as all the information provided in the literature, the simulation methodologies are also investigated at all levels. It can be concluded from the present review study that various factors, such as the uncertainty and repeatability of the brake wear experiments, distinguish the results of the subsystem and system levels. This gap should be taken into account in the development of future experimental and simulation methods for the investigation of airborne brake wear emissions.

Like gravitational wave detection, inter-spacecraft geodesy is a measurement of gravitational tidal accelerations deforming a constellation of two or more orbiting reference test masses (TM). The LISA TM system requires TM in free fall with residual stray accelerations approaching the fm/s${}^2$/Hz${}^{1/2}$ level in the mHz band, as demonstrated in the LISA Pathfinder “Einstein’s geodesic explorer” mission. Current geodesy missions are limited by accelerometers with 100 pm/s${}^2$/Hz${}^{1/2}$ level, due to intrinsic design limitations, as well as the challenging low Earth orbit environment and operating conditions. A reduction in the TM acceleration noise could lead to an important improvement in the scientific return of future geodesy missions focusing on mass change, especially in a scenario with multiple pairs of geodesy satellites. We present here a preliminary assessment of how the LISA TM system, known as the “gravitational reference sensor” (GRS), could be adapted for use in future geodesy missions aiming at residual TM accelerations noise at the pm/s${}^2$/Hz${}^{1/2}$ level, addressing the major design issues and performance limitations. We find that such a performance is possible in a geodesy GRS that is simpler and smaller than that used for LISA, with a lighter, sub-kg TM and gaps reduced from 4 mm to less than 1 mm. Acceleration noise performance limitations will likely be closely tied to the required levels of applied actuation forces on the TM.

This paper deals with (deep) neural network implementations of sensorimotor control for automated driving. We show how to construct complex behaviors by re-using elementary neural network building blocks that can be trained and tested extensively; one of our goals is to mitigate the “black box” and verifiability issues that affect end-to-end trained networks. By structuring complex behaviors within a subsumption architecture, we retain the ability to learn (mostly at motor primitives level) with the ability to create complex behaviors by subsuming the (well-known) learned elementary perception-action loops. The learning process itself is simplified, since the agent needs only to learn elementary behaviors. At the same time, the structure imposed with the subsumption architecture ensures that the agent behaves in predictable ways (e.g., treating all obstacles uniformly). We demonstrate these ideas for longitudinal obstacle avoidance behavior, but the proposed approach can also be adapted to other situations.