Scene flow estimation aims to predict 3D motion from consecutive point cloud frames, which is of great interest in autonomous driving field. Existing methods face challenges such as insufficient spatio-temporal modeling and inherent loss of fine-grained feature during voxelization. However, the success of Mamba, a representative state space model (SSM) that enables global modeling with linear complexity, provides a promising solution. In this paper, we propose MambaFlow, a novel scene flow estimation network with a mamba-based decoder. It enables deep interaction and coupling of spatio-temporal features using a well-designed backbone. Innovatively, we steer the global attention modeling of voxel-based features with point offset information using an efficient Mamba-based decoder, learning voxel-to-point patterns that are used to devoxelize shared voxel representations into point-wise features. To further enhance the model's generalization capabilities across diverse scenarios, we propose a novel scene-adaptive loss function that automatically adapts to different motion patterns. Extensive experiments on the Argoverse 2 benchmark demonstrate that MambaFlow achieves state-of-the-art performance with real-time inference speed among existing works, enabling accurate flow estimation in real-world urban scenarios. The code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/SCNU-RISLAB/MambaFlow</uri>.

The autonomous driving of heavy-duty tracked vehicles (HTVs) is prone to significant position and posture deviation at the end of walking part in trajectory tracking control, which may affect the efficiency and safety of vehicle operation seriously. Taking mining-used HTVs as an example, the driving environment is complex and harsh with numerous disturbances, making it difficult to conduct on-site experiments on autonomous driving. In this paper, a parallel mining HTVs autopilot system composed of artificial system and physical system is established based on the ACP (artificial system, computational experiment, parallel execution) method to solve the problem of difficult computational experiments of mining HTVs in the shaft roadway. The optimization of control strategies of miningused HTVs is completed through computational experiments in different scenarios. After the optimization of the strategies through computational experiments, the trajectory tracking effect was improved by 12.05% and 13.77% in two different scenarios, respectively; The average deviation angle optimization effect during operation reached 75.76% and 78.45% respectively, and the average deviation optimization effect reached 53.47% and 58.08%, respectively. Finally, bidirectional control and collaborative evolution were achieved through the interaction of virtual and real data between artificial system and physical system. The effectiveness of the control method was demonstrated by theoretical analysis and parallel execution tests.

Path-speed decomposition-based trajectory planning schemes have garnered widespread usage in real-world robotics applications due to their efficacy and computational efficiency. While a global route can be planned offline, generating a local path adaptive to real-time situations online remains essential. We propose a local path planning algorithm that prioritizes smoothness and low computational complexity, facilitating scalability to dense environments with various on-road entities. Our algorithm leverages a sparse graph structure to generate crucial obstacle-specific nodes and connect them via spline edges. Several conditional checks are introduced to maintain graph sparsity, boosting computational efficiency without compromising performance. The final path evaluation considers both the smoothness of the path and the risks to vulnerable road users. The effectiveness of the proposed algorithm is demonstrated through CARLA simulation studies and extensive comparative analysis against benchmarking methods. Finally, a scaled car demonstration with a dynamic vehicle on a curved road is presented to showcase the performance of the proposed method on a physical system.

In multi-agent real-time position management tasks, the accuracy of error convergence and convergence time are crucial. This paper reformulates the proposed real-time position management scheme as a quadratic programming problem with equality constraints and solves it in real-time using the zeroing neural dynamics (ZND) model. To enhance the model's ability to detect real-time position management errors, an adaptive parameter finite-time convergent zeroing neural dynamics (AP-FTZND) model is introduced, incorporating adaptive parameters and a nonlinear activation function (AF) within the ZND framework. The global convergence of the AP-FTZND model is proven using the Lyapunov theory, and the upper bound of the convergence time is derived. Finally, the effectiveness and superiority of the AP-FTZND model in solving multi-agent real-time position management tasks are validated through simulations and physical experiments.

This paper proposes a fault-tolerant system based on sensor fusion for the classification of detected objects to ensure the safety of autonomous driving. The system addresses and refines misclassifications made by black-box object detectors. Raw sensor data from camera, Light Detection And Ranging (LiDAR), and radar are extracted within the bounding box region identified by the black-box detector. The classification is based on the extracted images and point cloud data, with the inclusion of a ‘fault class’ to detect misclassifications by the black-box detectors. This approach also consolidates class labels to improve classification reliability and reduce computational costs. A dataset is built to evaluate the performance of system. The classification performance is analyzed based on different sensor types (camera, LiDAR, radar), machine learning techniques (attention mechanisms), and sensor fusion methods (early, middle, late). The classification result is obtained by probabilistically fusing the outputs of both black-box and fusion-based classifiers. The results demonstrate that this fusion-based approach significantly improves the reliability of the classification, providing a robust solution to safe autonomous driving.

Large language models (LLMs) as autonomous agents offer a novel avenue for tackling real-world challenges through a knowledge-driven manner. These LLM-enhanced methodologies excel in generalization and interpretability. However, the complexity of driving tasks often necessitates the collaboration of multiple, heterogeneous agents, underscoring the need for such LLM-driven agents to engage in cooperative knowledge sharing and cognitive synergy. Despite the promise of LLMs, current applications predominantly center around singleagent scenarios, which limits their scope in the face of intricate, interconnected tasks. To broaden the horizons of knowledgedriven strategies and bolster the generalization capabilities of autonomous agents, we propose the KoMA framework consisting of the multi-agent interaction, the multi-step planning, the shared-memory, and the ranking-based reflection modules to enhance multi-agents' decision-making in complex driving scenarios. Based on the framework's generated text descriptions of driving scenarios, the multi-agent interaction module enables LLM agents to analyze and infer the intentions of surrounding vehicles based on scene information, akin to human cognition. The multi-step planning module enables LLM agents to analyze and obtain final action decisions layer by layer to ensure consistent goals for short-term action decisions. The shared memory module can accumulate collective experience to make superior decisions, and the ranking-based reflection module can evaluate and improve agent behavior with the aim of enhancing driving safety and efficiency. The KoMA framework not only enhances the robustness and adaptability of autonomous driving agents but also significantly elevates their generalization capabilities across diverse scenarios. Empirical results demonstrate the superiority of our approach over traditional methods, particularly in its ability to handle complex, unpredictable driving environments without extensive retraining. Project Page: https://jkmhhh.github. io/KoMA/.

The recent progress in autonomous driving primarily relies on sensor-rich systems, encompassing radars, LiDARs, and advanced cameras, in order to perceive the environment. However, human-operated vehicles showcase an impressive ability to drive based solely on visual perception. This study introduces an end-to-end method for predicting the steering angle and vehicle speed exclusively from a monocular camera image. Alongside the color image, which conveys scene texture and appearance details, a monocular depth image and a semantic segmentation image are internally derived and incorporated, offering insights into spatial and semantic environmental structures. This results in a total of three input images. Moreover, LSTM units are also employed to acquire temporal features. The proposed model demonstrates a significant enhancement in RMSE compared to the state-of-the-art, achieving a notable improvement of 44.96% for the steering angle and 4.39% for the speed on the Udacity dataset. Furthermore, tests on the CARLA and Sully Chen datasets yield results that outperform those reported in the literature. Extensive ablation studies are also conducted to showcase the effectiveness of each component. These findings highlight the potential of self-driving systems using visual input alone.

Vehicles are sophisticated machines equipped with sensors that provide real-time data for onboard driving assistance systems. Due to the wide variety of traffic, road, and weather conditions, continuous system enhancements are essential. Connectivity allows vehicles to transmit previously unknown data, expanding datasets and accelerating the development of new data models. This enables faster identification and integration of novel data, improving system reliability and reducing time to market. Data Spaces aim to create a data-driven, interconnected, and innovative data economy, where edge and cloud infrastructures support a virtualised IoT platform that connects data sources and development servers. This paper proposes an edge-cloud data platform to connect car data producers with multiple and heterogeneous services, addressing key challenges in Data Spaces, such as data sovereignty, governance, interoperability, and privacy. The paper also evaluates the data platform's performance limits for text, image, and video data workloads, examines the impact of connectivity technologies, and assesses latencies. The results show that latencies drop to 33 ms with 5G connectivity when pipelining data to consuming applications hosted at the edge, compared to around 77 ms when crossing both edge and cloud processing infrastructures. The results offer guidance on the necessary processing assets to avoid bottlenecks in car data platforms.