Reasoning about human motion is an important prerequisite to safe and socially-aware robotic navigation. As a result, multi-agent behavior prediction has become a core component of modern human-robot interactive systems, such as self-driving cars. While there exist many methods for trajectory forecasting, most do not enforce dynamic constraints and do not account for environmental information (e.g., maps). Towards this end, we present Trajectron++, a modular, graph-structured recurrent model that forecasts the trajectories of a general number of diverse agents while incorporating agent dynamics and heterogeneous data (e.g., semantic maps). Trajectron++ is designed to be tightly integrated with robotic planning and control frameworks; for example, it can produce predictions that are optionally conditioned on ego-agent motion plans. We demonstrate its performance on several challenging real-world trajectory forecasting datasets, outperforming a wide array of state-ofthe-art deterministic and generative methods.
State-of-the-art autonomous driving systems rely on high precision map data. These map data are crucial to the driving function and therefore need to be validated during drive time. This work describes a probabilistic neural model inferring information about the road in front of an automated vehicle from sensory data. This problem is modeled as a pixelwise classification problem. Thereby, the limitations of systems relying on pre-processed map data are overcome by replacing navigation related map data with online sensor information. The proposed model is trained based on recorded driving traces and allows the facilitation of the vehicle odometry as input label. Two use cases, namely Path Planning and Map Validation are presented and evaluated.
It is a long-standing open question whether electrification of wind-blown sand due to tribocharging—the generation of electric charges on the surface of sand grains by particle–particle collisions—could affect rates of sand transport occurrence on Mars substantially. While previous wind tunnel experiments and numerical simulations addressed how particle trajectories may be affected by external electric fields, the effect of sand electrification remains uncertain. Here we show, by means of wind tunnel simulations under air pressure of 20 mbar, that the presence of electric charges on the particle surface can reduce the minimal threshold wind shear velocity for the initiation of sand transport, u
*ft, significantly. In our experiments, we considered different samples, a model system of glass beads as well as a Martian soil analog, and different scenarios of triboelectrification. Furthermore, we present a model to explain the values of u
*ft obtained in the wind tunnel that is based on inhomogeneously distributed surface charges. Our results imply that particle transport that subsides, once the wind shear velocity has fallen below the threshold for sustained transport, can more easily be restarted on Mars than previously thought.
Model Predictive Control (MPC) has become a popular framework in embedded control for high-performance autonomous systems. However, to achieve good control performance using MPC, an accurate dynamics model is key. To maintain real-time operation, the dynamics models used on embedded systems have been limited to simple first-principle models, which substantially limits their representative power. In contrast to such simple models, machine learning approaches such as neural networks have been shown to accurately model even complex dynamic effects, but their large computational complexity hindered combination with fast real-time iteration loops. With this work, we present Neural-MPC, a framework to efficiently integrate large, complex neural network architectures as dynamics models within a model-predictive control pipeline. Our experiments, performed in simulation and the real world on a highly agile quadrotor platform, demonstrate up to 83% reduction in positional tracking error when compared to stateof-the-art MPC approaches without neural network dynamics.
The growing number of intelligent components inside a car leads to a considerable increase in amount of the produced data. Context aware paradigm plays a major role in managing this data and offering a numerous number of prospects and advantages for existing and new intelligent applications inside the car. Following that, enabling context prediction promises reliable solutions in terms of enhancing the comfort of the occupants and vehicle dynamics. Moreover, this would be a great step toward facilitating highly automated and autonomous driving. However, due to the complex nature of the data resources in an intelligent car and also the lack of comprehensive studies on different aspects of this concept in automotive, defining a functional architecture for context prediction requires broad knowledge and better understanding of multiple domains which are involved and have impacts. In this paper, we investigate the most effective elements and factors in each one of the related domains which help to enable context prediction architectures inside the intelligent cars and analyze the feasible dimensions in detail, cover their advantages, and address the challenges ahead. We elucidate the possibility and validity of our considerations with the help of two use cases of adaptive HVAC and ACC systems.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.