Simultaneous trajectory prediction for multiple heterogeneous traffic participants is essential safe and efficient operation of connected automated vehicles under complex driving situations. Two main challenges this task are to handle the varying number target agents jointly consider factors that would affect their future motions. This because different kinds have motion patterns, behaviors affected by individual dynamics, interactions with surrounding agents, as well infrastructures. A...

Driving behavior modeling is of great importance for designing safe, smart, and personalized autonomous driving systems. In this paper, an internal reward function-based model that emulates the human's decision-making mechanism utilized. To infer function parameters from naturalistic human data, we propose a structural assumption about focuses on discrete latent intentions. It converts continuous problem to setting thus makes maximum entropy inverse reinforcement learning (IRL) tractable...

Deep reinforcement learning (DRL) is a promising way to achieve human-like autonomous driving. However, the low sample efficiency and difficulty of designing reward functions for DRL would hinder its applications in practice. In light this, this article proposes novel framework incorporate human prior knowledge DRL, order improve save effort sophisticated functions. Our consists three ingredients, namely, expert demonstration, policy derivation, RL. demonstration step, demonstrates their...

Due to its limited intelligence and abilities, machine learning is currently unable handle various situations thus cannot completely replace humans in real-world applications. Because exhibit robustness adaptability complex scenarios, it crucial introduce into the training loop of artificial (AI), leveraging human further advance algorithms. In this study, a real-time human-guidance-based (Hug)-deep reinforcement (DRL) method developed for policy an end-to-end autonomous driving case. With...

Predicting the behaviors of other agents on road is critical for autonomous driving to ensure safety and efficiency. However, challenging part how represent social interactions between output different possible trajectories with interpretability. In this paper, we introduce a neural prediction framework based Transformer structure model relationship among interacting extract attention target agent map waypoints. Specifically, organize into graph utilize multi-head encoder relations them. To...

To further improve learning efficiency and performance of reinforcement (RL), a novel uncertainty-aware model-based RL method is proposed validated in autonomous driving scenarios this paper. First, an action-conditioned ensemble model with the capability uncertainty assessment established as environment model. Then, developed based on adaptive truncation approach, providing virtual interactions between agent model, improving RL's performance. The then implemented end-to-end vehicle control...

Predicting the future states of surrounding traffic participants and planning a safe, smooth, socially compliant trajectory accordingly are crucial for autonomous vehicles (AVs). There two major issues with current driving system: prediction module is often separated from module, cost function hard to specify tune. To tackle these issues, we propose differentiable integrated (DIPP) framework that can also learn data. Specifically, our uses nonlinear optimizer as motion planner, which takes...

Autonomous driving systems require a comprehensive understanding and accurate prediction of the surrounding environment to facilitate informed decision-making in complex scenarios. Recent advances learning-based have highlighted importance integrating planning. However, this integration poses significant alignment challenges through consistency between patterns, interaction future To address these challenges, we introduce Hybrid-Prediction integrated Planning (HPP) framework, which operates...

This study aims to improve the performance and generalization capability of end-to-end autonomous driving with scene understanding leveraging deep learning multimodal sensor fusion techniques. The designed neural network takes as input visual image associated depth information in an early level outputs pixel-wise semantic segmentation vehicle control commands concurrently. learning-based model is tested high-fidelity simulated urban conditions compared benchmark CoRL2017 NoCrash. testing...

Optical coherence tomography (OCT) is susceptible to the coherent noise, which speckle noise that deteriorates contrast and detail structural information of OCT images, thus imposing significant limitations on diagnostic capability OCT. In this paper, we propose a novel image denoising method by using an end-to-end deep learning network with perceptually-sensitive loss function. The has been validated images acquired from healthy volunteers' eyes. label for training evaluating models are...

Energy management strategy (EMS) is very crucial for the hybrid power system of fuel cell range extender vehicles (FCREV) with respect to improving vehicle energy consumption economy. The objective this article formulate a real-time EMS efficiently regulate flow under complex driving conditions. First, nonlinear control developed calculate reference output current proton exchange membrane FC management, and theoretical stability proven by using Lyapunov function. Then, physics-based model...

With the aims of safe, smart and sustainable future mobility, a personalized approach trajectory planning control based on user preferences is developed for lane-change autonomous vehicles in this paper. First, safe area during lane change process identified by using constraint Delaunay triangulation. Then, an improved rapidly-exploring Random Trees (i-RRT) with B-spline to generate feasible cluster, which subject boundaries vehicle dynamics. To extract from we firstly adopt fuzzy linguistic...

Reinforcement learning (RL) requires skillful definition and remarkable computational efforts to solve optimization control problems, which could impair its prospect. Introducing human guidance into RL is a promising way improve performance. In this article, comprehensive guidance-based framework established. A novel prioritized experience replay mechanism that adapts in the process proposed boost efficiency performance of algorithm. To relieve heavy workload on participants, behavior model...

Learning-based approaches, such as reinforcement learning (RL) and imitation (IL), have indicated superiority over rule-based approaches in complex urban autonomous driving environments, showing great potential to make intelligent decisions. However, current RL IL still their own drawbacks, low data efficiency for poor generalization capability IL. In light of this, this paper proposes a novel learning-based method that combines deep from expert demonstrations, which is applied longitudinal...

Making safe and human-like decisions is an essential capability of autonomous driving systems, learning-based behavior planning presents a promising pathway toward achieving this objective. Distinguished from existing methods that directly output decisions, work introduces predictive framework learns to predict evaluate human data. This consists three components: generation module produces diverse set candidate behaviors in the form trajectory proposals, conditional motion prediction network...

Ensuring safety and achieving human-level driving performance remain challenges for autonomous vehicles, especially in safety-critical situations. As a key component of artificial intelligence, reinforcement learning is promising has shown great potential many complex tasks; however, its lack guarantees limits real-world applicability. Hence, further advancing learning, from the perspective, importance driving. revealed by cognitive neuroscientists, amygdala brain can elicit defensive...

Autonomous vehicles operating in complex real-world environments require accurate predictions of interactive behaviors between traffic participants. This paper tackles the interaction prediction problem by formulating it with hierarchical game theory and proposing GameFormer model for its implementation. The incorporates a Transformer encoder, which effectively models relationships scene elements, alongside novel decoder structure. At each decoding level, utilizes outcomes from previous...

Reinforcement learning (RL) is a promising approach in unmanned ground vehicles (UGVs) applications, but limited computing resource makes it challenging to deploy well-behaved RL strategy with sophisticated neural networks. Meanwhile, the training of on navigation tasks difficult, which requires carefully-designed reward function and large number interactions, yet can still fail due many corner cases. This shows intelligence current methods, thereby prompting us rethink combining human...

Decision-making for urban autonomous driving is challenging due to the stochastic nature of interactive traffic participants and complexity road structures. Although reinforcement learning (RL)-based decision-making schemes are promising handle scenarios, they suffer from low sample efficiency poor adaptability. In this paper, we propose Scene-Rep Transformer enhance RL capabilities through improved scene representation encoding sequential predictive latent distillation. Specifically, a...

The energy economy of fuel cell electric vehicles (FCEVs) plays a crucial role in determining their practicality, making the optimization management strategies (EMS) essential. Predictive EMS (PEMS) based on future vehicle speed prediction offers great potential for enhancing performance. However, current PEMS models rely historical data or static traffic information, overlooking impact real-time conditions. In this paper, we introduce Transformer-based (TPEMS) that incorporates predicted...

Owing to their extremely wide bandwidths, pure optical ultrasonic detection methods are gaining increasing interest. In this Letter, we proposed a simple detector that is based on the polarization-dependent reflection. When acoustic wave reaches liquid-glass interface, pressure changed relative refractive index between two media, leading perturbations in reflectance of probe beam glass. Unlike previous studies detected modulations intensity reflected beam, our method, named...

In this article, an improved short-term speed prediction method is proposed to predict future and analyze energy consumption of intelligent fuel cell vehicles. The predicted by the Inflated 3-D Inception long memory (LSTM) network, which takes spatiotemporal-vision information vehicle motion states. Specifically, spatiotemporal-vision-based deep neural network utilizes image sequences captured a front-facing camera as environmental historical series improve accuracy. Then, case study method,...

The performance of speed prediction-based energy management strategy (EMS) for fuel cell vehicles (FCVs) highly relies on the accuracy predicted sequences. Therefore, future sequences are estimated by Inflated 3D Inception long short-term memory (LSTM) network, which can use historical and image information to improve prediction. Meanwhile, economy powertrain system durability objectives real-time optimization. For optimizing FCVs, optimization EMS using LSTM network-based prediction is...