Retrieval Augmented Generation (RAG) with Knowledge Graphs (KGs) is an effective way to enhance Large Language Models (LLMs). Due to the natural discrepancy between structured KGs and sequential LLMs, KGs must be linearized to text before being inputted into LLMs, leading to the problem of KG Alignment with LLMs (KGA). However, recent KG+RAG methods only consider KGA as a simple step without comprehensive and in-depth explorations, leaving three essential problems unclear: (1) What are the factors and their effects in KGA? (2) How do LLMs understand KGs? (3) How to improve KG+RAG by KGA? To fill this gap, we conduct systematic explorations on KGA, where we first define the problem of KGA and subdivide it into the graph transformation phase (graph-to-graph) and the linearization phase (graph-to-text). In the graph transformation phase, we study graph features at the node, edge, and full graph levels from low to high granularity. In the linearization phase, we study factors on formats, orders, and templates from structural to token levels. We conduct substantial experiments on 15 typical LLMs and three common datasets. Our main findings include: (1) The centrality of the KG affects the final generation; formats have the greatest impact on KGA; orders are model-dependent, without an optimal order adapting for all models; the templates with special token separators are better. (2) LLMs understand KGs by a unique mechanism, different from processing natural sentences, and separators play an important role. (3) We achieved 7.3% average performance improvements on four common LLMs on the KGQA task by combining the optimal factors to enhance KGA.

In a dynamic environment, moving to the destination safely and effectively is of paramount importance for an unmanned ground vehicle (UGV). This article presents a strategy of trajectory planning and tracking that aims to ensure the UGV’s safety in an uncertain environment. Specifically, based on the initial environment information, a global optimal trajectory connecting the start and the destination is predefined by an artificial fish swarm algorithm (AFSA). In the presence of unforeseen obstacles, a trial-based forward search (TFS) algorithm based on the Markov chain is proposed in the local trajectory planning module, while collision prediction is integrated as heuristic information. The vehicle’s current state is updated accordingly for the sake of avoiding entire state spaces involved in the computation. Therefore, the storage efficiency and convergence rate in local path planning are sufficiently enhanced in comparison to dynamic programming. Moreover, command signals can be calculated with the proposed multiconstrained model predictive controller (MMPC), ensuring the vehicle to track the reference trajectory and smoothen the motion. Finally, the results in both simulations and experiments reveal the effectiveness of the proposed algorithm in the presence of both static and dynamic obstacles.Note to Practitioners—This article is motivated by the unmanned ground vehicle (UGV) collision avoidance problem in practical missions, such as farming and emergency response. In recent years, various trajectory planning and tracking algorithms have been widely developed. However, the environmental complexity and the intruders’ unexpected movement pose difficulties in trajectory planning, especially in ensuring the computation time under the allowable threshold. Moreover, the UGV practical trajectory tracking is a challenging task which demands a desired response within various physical constraints. In this article, a two-stage conflict resolution system is proposed. First, a trial-based forward search (TFS) is used to generate a new trajectory deviating the UGV from the initially generated trajectory by the artificial fish swarm algorithm (AFSA), aiming to avoid the unforeseen intruders (unknown in prior) in real-time. Using these two trajectory planning algorithms alternatively, both global trajectory optimality in a cluttered environment and appropriate maneuvers with respect to unexpected intruders can be achieved. Subsequently, the UGV is modeled according to its kinematic characteristics, and thus a multiconstrained model predictive controller (MMPC) is designed to follow the reference trajectory. The physical constraints are respected by integrating them into the controller. Simulations and experimental results demonstrate that the proposed strategy can guide and control a UGV from the start to the destination safely and smoothly, even in the case of multiple obstacles with constant or varying velocities. Furthermore, the proposed collision avoidance strategy can be extended to other unmanned systems, including unmanned aerial vehicles and unmanned surface vehicles.