Multi-agent trajectory prediction plays an increasingly critical role in intelligent transportation systems. Despite significant progress in this field, several key challenges remain unresolved. Future trajectories of agents are jointly influenced by individual behavior patterns and the surrounding environment, while most existing methods extract temporal features at a single time scale, limiting their capacity to capture complex temporal dependencies within trajectory sequences. Moreover, many current approaches employ numerically precise formulations for interaction modeling, which are misaligned with the imprecise nature of real-world social behavior. To address these limitations, we propose a multi-agent trajectory prediction model that combines multi-scale temporal encoding and egocentric scan-based spatial representation. Temporally, we leverage a sliding-window-based multi-scale temporal encoder to capture trajectory features across diverse time scales. Spatially, we partition the surrounding environment into multiple egocentric bins to represent social zones, thereby simulating real-world interaction patterns. Experimental results on public benchmark datasets ETH/UCY and SDD demonstrate that our model outperforms existing approaches.

This letter presents a sample-efficient deep reinforcement learning (DRL) based method to address the intricate task of autonomous navigation in dynamic, constrained environments. The proposed method leverages a graph neural network with feature-wise linear modulation modules to effectively extract features from observations modeled as heterogeneous graph. By transferring offline model predictive control (MPC) experience data, which can be generated fully in parallel, the training process can be jump-started to cope with the sparse rewards. Simulation results demonstrate a 98% navigation success rate, surpassing baselines by at least 6%, while halving the training steps. Furthermore, we successfully deploy our policy network on a real robot and contribute our code to the community through open-sourcing: github.com/TIB-K330/drl_planner.

Some large social environments are expected to use Covered Path Planning (CPP) methods to handle daily tasks such as cleaning and disinfection. These environments are usually large in scale, chaotic in structure, and contain many obstacles. The proposed method is based on the improved SCAN-STC (Spanning Tree Coverage) method and significantly reduces the solution time by optimizing the backtracking module of the algorithm. The proposed method innovatively introduces the concept of optimal backtracking points to sacrifice the spatial complexity of the algorithm to reduce its computational complexity. The necessity of backtracking in such environments is proved to illustrate the generalization ability of the method. Finally, based on secondary coding, the STC solution is explicitly expressed as a continuous and cuttable global path, which can be generalized to Multi-robot Covered Path Planning (MCPP) to avoid the path conflict problem in the multi-robot system, and the paths assigned to each robot have good balance. The method of this study is proven to be effective through simulations in various random environments and a real environment example. Compared with the advanced methods, the computational time is reduced by 82.47%.

Analyzing and forecasting trajectories of agents like pedestrians plays a pivotal role for embodied intelligent applications. The inherent indeterminacy of human behavior and complex social interaction among a rich variety of agents makes this task more challenging than common time-series forecasting. In this letter, we aim to explore a distinct formulation for multi-agent trajectory prediction framework. Specifically, we propose a patching-based temporal feature extraction module and a graph-based social feature extraction module, enabling effective feature extraction and cross-scenario generalization. Moreover, we reassess the role of social interaction and present a novel method based on explicit modality modulation to integrate temporal and social features, thereby constructing an efficient single-stage inference pipeline. Results on public benchmark datasets demonstrate the superior performance of our model compared with the state-of-the-art methods. The code is available at: https://github.com/TIB-K330/pmm-net.