Search Results

Urea/SCR systems have been proven effective at reducing NOx over a wide range of operating conditions on mid/heavy duty diesel vehicles. However, design changes due to reduction in the size of modern compact Urea/SCR systems and lower exhaust temperature have increased the possibility of urea deposit formation. Urea deposits are formed when urea in films and droplets undergoes undesirable secondary reactions and generate by-products such as ammelide, biuret and cyanuric Acid (CYA). Ammelide and CYA are difficult to decompose which lead to the formation of solid deposits on the surface. This phenomenon degrades the performance of the after treatment system by decreasing overall mixing efficiency, lowering de-NOx efficiency and increasing pressure drop. Therefore, mitigating urea deposits is a primary design goal of modern diesel after-treatment systems.

In this paper, a novel online coverage path planning (CPP) method for autonomous sweeper trucks in closed areas is proposed. This method can efficiently generate executable paths for sweeper trucks that cover all feasible uncleaned areas without getting tracked in dead-end, i.e., no backward behaviors required and avoid dynamic obstacles. To reach that end, a modified biological inspired neuron network method considering vehicle constrains is developed, where the dynamic of each neuron is determined by the shunting function. The path will be iteratively generated based on local neuron dynamics. In order to avoid dead-end, a detour algorithm combing with back iteration is introduced to search the nearest uncleaned area that can be reached within vehicle constrains. The proposed method is empirically approved to be computationally efficient and adaptive to maps with arbitrary shapes.

Exploring an unknown place autonomously is a challenge for robots, especially when the environment is changing. Moreover, in real world application, efficient path planning is crucial for autonomous vehicles to have timely response to execute a collision-free motion. In this paper we focus on environment exploration which enables an automated system to establish a map of an unknown environment with unforeseen objects moving within it. We introduce an exploration package that enables robots self-exploration with an online collision avoidance planner. The package consists of exploration module, global planner module and local planner module. We modularize the package so that developers can easily make modifications or even substitutions to some modules for their specific application. In order to validate the algorithm, we designed and built a robot car as a low cost validation platform to test the autonomous vehicle algorithms in the real world.

Global deployment of autonomous capability for commercial vehicles is a big challenge. In order to improve the robustness of autonomous approach under different traffic scenarios, environments, road conditions, and driver behaviors, a combined approach of virtual simulation, vehicle-in-the-loop (VIL) testing, proving ground testing, and final field testing have been established for algorithms validation. During the validation platform setup, different platforms for different functionalities have been studied, including open source virtual testing environment (CARLA, AirSim), and commercial one (IPG). We also cooperate with MCity to do proving ground validation. In virtual testing, the functionality of sensors (camera, radar, Lidar, GPS, IMU) and vehicle dynamic models can be applied in the virtual environment. In VIL testing, real world and virtual test will be connected for different validation purposes.

For commercial-vehicle Original Equipment Manufacturers (OEMs), predictive maintenance has drawn attention for the benefits of money saving and increased road safety. Data-driven models have been widely explored and implemented as predictive maintenance solutions. However, the working scenarios for different commercial-vehicles vary a lot, which makes it difficult to build a universal model suitable for all the cases. In this paper, we propose a Recurrent Neural Adaptive Processes (RNAPs) network to adapt to different scenarios by modeling the uncertain at the same time. The ensemble network combines the traits of neural processes, recurrent neural network and meta learning together. Neural processes consider the context information to calculate the uncertainty and improve the prediction results. Meta-learning works well when dealing with few-shot multi-tasks learning, and recurrent networks are utilized as the encoder of the proposed model to process time-series data.