Search

Recent advances in monocular 3D detection leverage a depth estimation network explicitly as an intermediate stage of the 3D detection network. Depth map approaches yield more accurate depth to objects than other methods thanks to the depth estimation network trained on a large-scale dataset. However, depth map approaches can be limited by the accuracy of the depth map, and sequentially using two separated networks for depth estimation and 3D detection significantly increases computation cost and inference time. In this work, we propose a method to boost the RGB image-based 3D detector by jointly training the detection network with a depth prediction loss analogous to the depth estimation task. In this way, our 3D detection network can be supervised by more depth supervision from raw LiDAR points, which does not require any human annotation cost, to estimate accurate depth without explicitly predicting the depth map. Our novel object-centric depth prediction loss focuses on depth around foreground objects, which is important for 3D object detection, to leverage pixel-wise depth supervision in an object-centric manner. Our depth regression model is further trained to predict the uncertainty of depth to represent the 3D confidence of objects. To effectively train the 3D detector with raw LiDAR points and to enable end-to-end training, we revisit the regression target of 3D objects and design a network architecture. Extensive experiments on KITTI and nuScenes benchmarks show that our method can significantly boost the monocular image-based 3D detector to outperform depth map approaches while maintaining the real-time inference speed., Comment: Accepted by IEEE Transaction on Intelligent Transportation System (T-ITS)

Recent years have seen the rapid development of autonomous driving systems, which are typically designed in a hierarchical architecture or an end-to-end architecture. The hierarchical architecture is always complicated and hard to design, while the end-to-end architecture is more promising due to its simple structure. This paper puts forward an end-to-end autonomous driving method through a deep reinforcement learning algorithm Dueling Double Deep Q-Network, making it possible for the vehicle to learn end-to-end driving by itself. This paper firstly proposes an architecture for the end-to-end lane-keeping task. Unlike the traditional image-only state space, the presented state space is composed of both camera images and vehicle motion information. Then corresponding dueling neural network structure is introduced, which reduces the variance and improves sampling efficiency. Thirdly, the proposed method is applied to The Open Racing Car Simulator (TORCS) to demonstrate its great performance, where it surpasses human drivers. Finally, the saliency map of the neural network is visualized, which indicates the trained network drives by observing the lane lines. A video for the presented work is available online,https://youtu.be/76ciJmIHMD8orhttps://v.youku.com/v_show/id_XNDM4ODc0MTM4NA==.html.

Prior arts in the field of motion predictions for autonomous driving tend to focus on finding a trajectory that is close to the ground truth trajectory. Such problem formulations and approaches, however, frequently lead to loss of diversity and biased trajectory predictions. Therefore, they are unsuitable for real-world autonomous driving where diverse and road-dependent multimodal trajectory predictions are critical for safety. To this end, this study proposes a novel loss function, \textit{Lane Loss}, that ensures map-adaptive diversity and accommodates geometric constraints. A two-stage trajectory prediction architecture with a novel trajectory candidate proposal module, \textit{Trajectory Prediction Attention (TPA)}, is trained with Lane Loss encourages multiple trajectories to be diversely distributed, covering feasible maneuvers in a map-aware manner. Furthermore, considering that the existing trajectory performance metrics are focusing on evaluating the accuracy based on the ground truth future trajectory, a quantitative evaluation metric is also suggested to evaluate the diversity of predicted multiple trajectories. The experiments performed on the Argoverse dataset show that the proposed method significantly improves the diversity of the predicted trajectories without sacrificing the prediction accuracy., RA-L accepted

There has been an increasing level of demand for faster, safer and greener transportation systems with higher levels of capacity and convenience, though the implementation of transportation systems overall is often restricted by geographical limitations, presenting a challenge to scientists and engineers in the field. However, we have been witnessing the evolution of the transportation systems over the last few decades, and at present we are facing a new era of intelligent transportation systems (ITS) empowered by artificial intelligence (AI) technologies. There have been classification, deep learning, and reinforcement learning techniques, to name a few, which collectively have enabled almost all technical elements of the ITS. For example, autonomous vehicle technologies are now mature enough to introduce self-driving cars, taxis, buses, and trucks on the roads and streets; traffic signals are controlled by AI-based systems for far more enhanced traffic efficiency; and machine learning based on big data is improving the operational performance of transportation systems to the next level of safety, efficiency, and sustainability.

In practice, many autonomous vehicle developers put a tremendous amount of time and efforts in tuning and calibrating the path tracking controllers in order to achieve robust tracking performance and smooth steering actions over a wide range of vehicle speed and road curvature changes. This design process becomes tiresome when the target vehicle changes frequently. In this study, a model-based Linear Quadratic Gaussian (LQG) Control with adaptive Q-matrix is proposed to efficiently and systematically design the path tracking controller for any given target vehicle while effectively handling the noise and error problems arise from the localization and path planning algorithms. The regulator, in turn, is automatically designed, without additional efforts for tuning at various speeds. The performance of the proposed algorithm is validated based on KAIST autonomous vehicle. The experimental results show that the proposed LQG with adaptive Q-matrix has tracking performance in both low (15kph) and high (45kph) speed driving conditions better than those of other conventional tracking methods like the Stanley and Pure-pursuit methods.

Powertrain configurations described with elementary (physical) levers can intuitively depict the connections between planetary gear (PG) shafts and powertrain components. However, finding optimal compound-split hybrid configurations using the elementary lever is practically impossible due to the large design space. In fact, each of the existing 252 compound-split configurations has three design variables: two PG gear ratios and a final drive gear ratio. In this paper, a compound (virtual) lever-based design methodology that eliminates redundant elementary lever designs is proposed to enable a full compound-split hybrid electric vehicles design domain search. The performance metrics were assessed in the compound lever design space. Later, the designs were converted back to elementary lever configurations by applying a design space conversion map, and their performance metrics were plotted on a fuel economy versus acceleration performance plane to compare the potential of the 252 compound-split configurations. Finally, an optimal configuration that can reach 0-160 kph in 15.36 sec, which is 5.90 sec faster than that of the Prius configuration, while maintaining a competitive fuel economy, was selected. The proposed method revealed that there are still many configurations that are potentially better than the commercially available split hybrids.

In this paper, we address the problem of predicting the future motion of a dynamic agent (called a target agent) given its current and past states as well as the information on its environment. It is paramount to develop a prediction model that can exploit the contextual information in both static and dynamic environments surrounding the target agent and generate diverse trajectory samples that are meaningful in a traffic context. We propose a novel prediction model, referred to as the lane-aware prediction (LaPred) network, which uses the instance-level lane entities extracted from a semantic map to predict the multi-modal future trajectories. For each lane candidate found in the neighborhood of the target agent, LaPred extracts the joint features relating the lane and the trajectories of the neighboring agents. Then, the features for all lane candidates are fused with the attention weights learned through a self-supervised learning task that identifies the lane candidate likely to be followed by the target agent. Using the instance-level lane information, LaPred can produce the trajectories compliant with the surroundings better than 2D raster image-based methods and generate the diverse future trajectories given multiple lane candidates. The experiments conducted on the public nuScenes dataset and Argoverse dataset demonstrate that the proposed LaPred method significantly outperforms the existing prediction models, achieving state-of-the-art performance in the benchmarks., Comment: 13 pages, 2 figures, 7 tables, CVPR 2021

Predicting the future trajectory of surrounding vehicles in a randomly varying traffic level is one of the most challenging problems in developing an autonomous vehicle. Since there is no pre-defined number of interacting vehicles participate in, the prediction network has to be scalable with respect to the vehicle number in order to guarantee the consistency in terms of both accuracy and computational load. In this paper, the first fully scalable trajectory prediction network, SCALE-Net, is proposed that can ensure both higher prediction performance and consistent computational load regardless of the number of surrounding vehicles. The SCALE-Net employs the Edge-enhance Graph Convolutional Neural Network (EGCN) for the inter-vehicular interaction embedding network. Since the proposed EGCN is inherently scalable with respect to the graph node (an agent in this study), the model can be operated independently from the total number of vehicles considered. We evaluated the scalability of the SCALE-Net on the publically available NGSIM datasets by comparing variations on computation time and prediction accuracy per single driving scene with respect to the varying vehicle number. The experimental test shows that both computation time and prediction performance of the SCALE-Net consistently outperform those of previous models regardless of the level of traffic complexities., 8 pages, 9 figures, and 2 tables, the paper is submitted to the conference IROS2020 and is under review

The recent rise of renewable energy technologies in the building sector is expected to reduce fossil fuel consumption but leads to a greater complexity in the design and control of Heating, Ventilation, and Air Conditioning (HVAC) systems. As a consequence, the traditional control approach does not fully exploit the potential of the photovoltaic-assisted hybrid HVACs. This paper presents an investigation on the energy cost savings and the optimal control strategies of a photovoltaic-assisted hybrid heating system based on optimal control theory. The considered system consists of a radiant floor heating system, a gas boiler and a photovoltaic-assisted air-source heat pump (AS-HP) as heat sources, with a water tank as thermal energy storage (TES). The building thermal dynamics and all the components of the heating system were modelled in MATLAB environment together with a baseline rule-based controller (RBC). The optimal control problem is formulated such that cost function, constraints, state and control variables are defined. Due to a large number of states and control variables, the optimal control problem is converted to the nonlinear parameter optimization problem, and the solution is obtained by using nonlinear programming (NLP). For simulation settings, model parameters, weather, and energy demand profiles were adopted from historical data from a North Italian case study. The simulation results show that the photovoltaic-assisted hybrid heating system coupled with optimal energy management strategy can potentially save up to 20% of the energy consumption cost when compared to the state-of-art RBC and increase the photovoltaic self-consumption by 30%.

Previously, an equivalent consumption minimization strategy (ECMS) was developed that provides near-optimal performance of hybrid vehicles based on an adaptation of equivalence factor from state of charge feedback. However, under real-world driving conditions with uncertainties, such as hilly roads, ECMS requires a predictive scheme utilizing future driving information in order to prevent a loss of optimality. In this paper, we synthesize predictive ECMS in a feedforward way to adjust the equivalence factor based on its theoretical connection with future driving statistics, in a systematic manner. First, a useful noncausal adaptation strategy is extracted from dynamic programming results. Then, the inverse problem is formulated and solved to derive an explicit representation of the constant optimal equivalence factor with justified assumptions. Finally, a causal, predictive adaptation strategy using this closed-form solution is synthesized to mimic the noncausal one, and its effectiveness is evaluated for fuel cell hybrid electric vehicles. Results show that if the predicted statistical information reflects well the future driving conditions, the proposed strategy accurately estimates the constant optimal equivalence factor, including the jump behavior, thereby yielding less than 1.5% loss of fuel optimality. Moreover, this approach is extendible to other configurations.

Platoon formation of highway vehicles has the potential to significantly enhance road safety, improve highway utility, and increase traffic efficiency. However, various uncertainties and disturbances that are present in real-world driving conditions make the implementation of vehicular platoon a challenging problem. This study presents an H-infinity control method for a platoon of heterogeneous vehicles with uncertain vehicle dynamics and uniform communication delay. The requirements of string stability, robustness and tracking performance are systematically measured by the H-infinity norm, and explicitly satisfied by casting into the linear fractional transformation format. A delay-dependent linear matrix inequality is derived to numerically solve the distributed controllers for each vehicle. The performances of the controlled platoon are theoretically analysed by using a delay-dependent Lyapunov function which includes a linear quadratic function of states during the delay period. Simulations with a platoon of heterogeneous vehicles are conducted to demonstrate the effectiveness of the proposed method under random parameters and external disturbances.

A wireless charging or inductive charging electric vehicle (EV) is a type of EVs with a battery that is charged from a charging infrastructure, using a wireless power transfer technology. Wireless charging EVs are classified as stationary or dynamic charging EVs. Stationary charging EVs charge wirelessly when they are parked, and dynamic charging EVs can charge while they are in motion. The online electric vehicle developed at the Korea Advanced Institute of Science and Technology is an example of a commercially available dynamic charging transportation system. Numerous studies have reported that one of the benefits of dynamic charging is that it allows smaller and lighter batteries to be used, due to frequent charging using the charging infrastructure embedded under roads. In this paper, we quantitatively analyze the benefits of dynamic charging with an economic model of battery size and charging infrastructure allocation, using a mathematical optimization model. Particularly, we analyze by how much battery size can be reduced and what the cost saving of reducing the battery size is with the model. We also show that the dynamic charging can be beneficial to battery life.

Global warming, air pollution, and fuel depletion have accelerated the deployment of hybrid electric vehicles (HEVs). Apart from the energy management, the configuration of hybrid powertrains plays a central role in achieving better fuel economy and enhanced drivability. This paper comparatively summarizes the configurations, modeling, and optimization techniques of HEVs. Four types of hybrid powertrain configurations available in the market, i.e., series, parallel, power-split and multi-mode, are introduced firstly, followed by their state-of-the-art and pros/cons. Among all configurations, multi-mode hybrid powertrains are observed to have the potential for utilizing the benefits of the other three types by switching the operating modes. Subsequently, the configuration generation and modeling techniques are summarized. By adopting the automated modeling method, the entire design space can be explored exhaustively, and 14 feasible configuration types are classified based on the binary tree. Finally, the research gaps and future trends of HEV configuration studies are discussed.

A sophisticated BMS often requires an accurate yet simple battery model. Simplified models such as the single particle model (SPM), however, provide limited insight because lithium concentration variations over the cell thickness dimension are often neglected. In this study, we propose a simplified electrochemical model that, by reviving the lithium concentration variation across the cell thickness, provides significantly improved prediction power over the SPM. The lithium concentration profiles are first derived according to the electrochemistry-based pseudo-two dimensions (P2D) model under the steady-state assumption, they are then relaxed into a dynamic model. By employing steady-state concentration profiles coupled with averaged dynamics, the proposed model predicts electrolyte and solid surface concentrations (ce and css, respectively) and potential variations across the cell during both transients and stead-state conditions. The proposed model is validated by comparing the cell voltage, ce, and css concentration RMS errors with respect to the P2D model under pulse and constant current inputs. The simulation shows that the proposed model achieves at least 74%, 77%, and 65% RMS error reductions for cell voltage, css and ce, respectively, when compared to SPM in constant current simulation. Finally, the presented model is used to predict and detect local lithium plating, which the SPM is not capable of doing; our method yields predictions similar to those of the original P2D model at computational loads comparable to SPM.

Among existing energy management strategies (EMSs) for fuel cell hybrid electric vehicles (FCHEV), the equivalent consumption minimization strategy (ECMS) is often considered as a practical approach because it can be implemented in real-time, while achieving near-optimal performance. However, under real-world driving conditions with uncertainties such as hilly roads, both near-optimality and charge-sustenance of ECMS are not guaranteed unless the equivalent factor (EF) is optimally adjusted in real-time. In this paper, a methodology of extracting the globally optimal EF trajectory from dynamic programming (DP) solution is proposed for the design of EF adaptation strategies. In order to illustrate the performance and process of the extraction method, a FCHEV energy management problem under hilly road conditions is investigated as a case study. The main goal is to learn how EF should be adjusted and the impact of EF adaptation on fuel economy under several hilly road cases. Using the extraction method, the DP-based EF is computed, and its performance is compared with those of Pontryagin's minimum principle (PMP) and conventional ECMS. The results show that the optimal EF adaptation significantly improves fuel economy when the battery SoC constraint becomes active, and thus EF must be properly adjusted under severely hilly road conditions.

Most of the prior studies on power-split hybrid electric vehicle's (PS-HEV) design focused on the powertrain configuration optimization. Yet, depicting the selected configuration is highly required for further design steps, ultimately manufacturing. This paper proposes an automatic approach to generate all the feasible kinematic diagrams for a given configuration with a single planetary gear (PG) set. While the powertrain configuration, which is the output of prior studies, illustrates the connection of the powertrain components to the PG, the kinematic diagram is a schematic diagram depicting the connections and arrangements of the components. First, positioning diagrams, specifying the position of the components with respect to each other and to the PG, are used to find all the possible arrangements. Then, given that the positioning diagrams have a one-to-one relationship with the kinematic diagrams, the feasible kinematic diagrams are identified using a set of feasibility rules applicable to the positioning diagrams. Finally, few guidelines are introduced to select good kinematic diagrams that best suit the overall vehicle design. Various configurations were investigated, and three of them including Prius and Voltec first-generation single PG configurations are discussed. The study reveals that the kinematic diagrams that have been patented are only a subset of all the feasible kinematic diagrams, and that even some good kinematic diagrams with better manufacturability are identified using this methodology. Thus, this methodology guarantees the search of the entire design space and the selection of kinematic diagrams that best suit the desired vehicle.

The majority of hybrid electric vehicles (HEVs) available on the market are power-split hybrid vehicles with a single planetary gear (PG), including the popular Toyota Prius and Chevy Volt. Although both vehicles use a single PG, they have different configurations with different numbers of operating modes. The Prius has no clutch and has a single operating mode, whereas the Chevy Volt uses three clutches and has four modes. The goal of this paper is to present a thorough analysis of all possible configurations of power-split hybrid powertrain using a single PG. The analysis includes the following steps: 1) search for all possible ways to connect powertrain elements to the PG; 2) identify all potential locations for clutch installations around the PG and examine the feasibility of additional operating modes introduced by the clutch installation; and 3) optimize fuel economy for performance comparison. The proposed analysis shows that a single PG can produce 12 different configurations, each of which can have four feasible operating modes by adding three clutches to the PG. In case studies, we focus on the two configurations that are used in the Prius and Volt to find the impact of adding (or removing) clutches and modes on their fuel economy performance. Our results show that adding one clutch to the Prius transmission (which is named “Prius+”) can significantly improve fuel economy in urban driving, whereas removing two clutches from the Volt transmission (“Volt-”) will not significantly affect fuel economy in both urban and highway driving. This multimode configuration analysis can be used to systematically design future power-split HEVs.

Under hilly road conditions, it is difficult to achieve near-optimal performance of energy management strategy (EMS) of fuel cell hybrid electric vehicle (FCHEV). In order to achieve near-optimality, optimal state reference trajectory is predicted based on future information, and thus reference tracking controller is often considered as real-time predictive EMS. There are two approaches depending on in what way the predicted reference will be used as follows: 1) position-based predictive EMS for tracking position- dependent reference, 2) time-based predictive EMS for tracking time-dependent reference. In this paper, analytical sensitivity analysis based on Pontryagin’s minimum principle (PMP) is performed to prove robustness of position-based predictive EMS with respect to velocity uncertainty. First, optimal control problem is formulated in time and position domain, and PMP approach is used to derive boundary value problem (BVP) that achieves global optimality. Then, sensitivity differential equations are developed which describe sensitivity of original BVP with respect to velocity uncertainty. Finally, these equations will be solved simultaneously with the original BVP to compute first-order sensitivity of time- and position- dependent optimal state. Results show that sensitivity of time-dependent optimal state is much bigger than that of position-dependent optimal state because velocity uncertainty can change predicted travel time, and this effect on sensitivity is significant. Therefore, predictive EMS should use current position to track position-dependent optimal state reference in terms of the robustness with respect to velocity uncertainty.