Pedestrian trajectory forecasting is crucial in various applications such as autonomous driving and mobile robot navigation. In such applications, camera-based perception enables the extraction of additional modalities (human pose, text) to enhance prediction accuracy. Indeed, we find that textual descriptions play a crucial role in integrating additional modalities into a unified understanding. However, online extraction of text requires the use of VLM, which may not be feasible for resource-constrained systems. To address this challenge, we propose a multimodal knowledge distillation framework: a student model with limited modality is distilled from a teacher model trained with full range of modalities. The comprehensive knowledge of a teacher model trained with trajectory, human pose, and text is distilled into a student model using only trajectory or human pose as a sole supplement. In doing so, we separately distill the core locomotion insights from intra-agent multi-modality and inter-agent interaction. Our generalizable framework is validated with two state-of-the-art models across three datasets on both ego-view (JRDB, SIT) and BEV-view (ETH/UCY) setups, utilizing both annotated and VLM-generated text captions. Distilled student models show consistent improvement in all prediction metrics for both full and instantaneous observations, improving up to ∼13%. The code is available at github.com/Jaewoo97/KDTF.

Motion planning is a crucial component of autonomous robot driving. While various trajectory datasets exist, effectively utilizing them for a target domain remains challenging due to differences in agent interactions and environmental characteristics. Conventional approaches, such as domain adaptation or ensemble learning, leverage multiple source datasets but suffer from domain imbalance, catastrophic forgetting, and high computational costs. To address these challenges, we propose Interaction-Merged Motion Planning (IMMP), a novel approach that leverages parameter checkpoints trained on different domains during adaptation to the target domain. IMMP follows a two-step process: pre-merging to capture agent behaviors and interactions, sufficiently extracting diverse information from the source domain, followed by merging to construct an adaptable model that efficiently transfers diverse interactions to the target domain. Our method is evaluated on various planning benchmarks and models, demonstrating superior performance compared to conventional approaches.

Safe and effective motion planning is crucial for autonomous robots. Diffusion models excel at capturing complex agent interactions, a fundamental aspect of decision-making in dynamic environments. Recent studies have successfully applied diffusion models to motion planning, demonstrating their competence in handling complex scenarios and accurately predicting multi-modal future trajectories. Despite their effectiveness, diffusion models have limitations in training objectives, as they approximate data distributions rather than explicitly capturing the underlying decision-making dynamics. However, the crux of motion planning lies in non-differentiable downstream objectives, such as safety (collision avoidance) and effectiveness (goal-reaching), which conventional learning algorithms cannot directly optimize. In this paper, we propose a reinforcement learning-based training scheme for diffusion motion planning models, enabling them to effectively learn non-differentiable objectives that explicitly measure safety and effectiveness. Specifically, we introduce a reward-weighted dynamic thresholding algorithm to shape a dense reward signal, facilitating more effective training and outperforming models trained with differentiable objectives. State-of-the-art performance on pedestrian datasets (CrowdNav, ETH-UCY) compared to various baselines demonstrates the versatility of our approach for safe and effective motion planning.

End-to-end autonomous driving (E2E-AD) has emerged as a promising paradigm that unifies perception, prediction, and planning into a holistic, data-driven framework. However, achieving robustness to varying camera viewpoints, a common real-world challenge due to diverse vehicle configurations, remains an open problem. In this work, we propose VR-Drive, a novel E2E-AD framework that addresses viewpoint generalization by jointly learning 3D scene reconstruction as an auxiliary task to enable planning-aware view synthesis. Unlike prior scene-specific synthesis approaches, VR-Drive adopts a feed-forward inference strategy that supports online training-time augmentation from sparse views without additional annotations. To further improve viewpoint consistency, we introduce a viewpoint-mixed memory bank that facilitates temporal interaction across multiple viewpoints and a viewpoint-consistent distillation strategy that transfers knowledge from original to synthesized views. Trained in a fully end-to-end manner, VR-Drive effectively mitigates synthesis-induced noise and improves planning under viewpoint shifts. In addition, we release a new benchmark dataset to evaluate E2E-AD performance under novel camera viewpoints, enabling comprehensive analysis. Our results demonstrate that VR-Drive is a scalable and robust solution for the real-world deployment of end-to-end autonomous driving systems.

In cloud storage applications where the data is shared by multiple mobile users, it is essential to provide the consistency among mobile users by means of appropriate synchronization algorithms. In particular, if the data is frequently updated and the number of mobile users sharing the data is large, the synchronization traffic can be significant. Moreover, the excessive synchronization traffic in mobile networks is more important in terms of radio resource utilization and energy consumption. In this paper, we propose an efficient delta synchronization (EDS) algorithm that aggregates the updated data to reduce the synchronization traffic and synchronizes the aggregated one periodically to satisfy the consistency. To find out the optimal policy for the aggregation and the periodical synchronization, an optimization problem is formulated as a Markov decision process (MDP) and a value iteration algorithm is presented for computing the stationary deterministic policy. Numerical results demonstrate that EDS can choose the optimal action that strikes a balance between the reduction of the synchronization traffic and the satisfaction of the consistency.

Distributed mobility management (DMM) is a promising technology to address the mobile data traffic explosion problem. Since the location information of mobile nodes (MNs) are distributed in several mobility agents (MAs), DMM requires an additional mechanism to share the location information of MNs between MAs. In the literature, multicast or distributed hash table (DHT)-based sharing methods have been suggested; however they incur significant signaling overhead owing to unnecessary location information updates under frequent handovers. To reduce the signaling overhead, we propose a timer-based Bloom filter aggregation (TBFA) scheme for distributing the location information. In the TBFA scheme, the location information of MNs is maintained by Bloom filters at each MA. Also, since the propagation of the whole Bloom filter for every MN movement leads to high signaling overhead, each MA only propagates changed indexes in the Bloom filter when a pre-defined timer expires. To verify the performance of the TBFA scheme, we develop analytical models on the signaling overhead and the latency and devise an algorithm to select an appropriate timer value. Extensive simulation results are given to show the accuracy of analytical models and effectiveness of the TBFA scheme over the existing DMM scheme.

Current evolved packet core (EPC) systems in LTE/LTE-A networks suffer from the exponentially increased mobile traffic and thus research on new EPC architectures is ongoing. In this paper, we introduce a virtualized EPC (vEPC) system where middlebox functions of the existing EPC systems are implemented in virtualized software modules and the virtualized software modules operate over selected physical service nodes. To minimize the impact of the increased transmission cost in vEPC, the optimal placement of middlebox functions is also investigated. Numerical results demonstrate that the proposed scheme can achieve lower packet transmission cost than the conventional one.

Recently, a data packet forwarding scheme between evolved node Bs (eNBs) in long-term evolution (LTE)/LTE-advanced (LTE-A) networks has been proposed to reduce the signaling overhead and delay incurred in the data path switching scheme. However, the conventional data packet forwarding scheme suffers from the increased delay when the length of the data packet forwarding chain is inappropriately long. To attain the optimal handover performance in LTE/LTE-A networks, we propose an optimized and distributed data packet forwarding scheme where an optimal length of the forwarding chain is obtained by a Markov decision process (MDP). Numerical results demonstrate that the proposed scheme achieves the optimal and adaptive performance in diverse network environments.

In wireless data access applications, reduction of both the access latency and the wireless traffic volume is essential. In this paper, we propose a fast wireless data access scheme (FW-DAS) for wireless data access applications in which data objects are frequently updated and fast access to data objects is indispensable. In FW-DAS, different operation modes are defined depending on the data object popularity, and only popular data objects are proactively pushed to the access point/base station to minimize the access latency while mitigating the traffic load over the wireless link. An analytical model for the access latency is developed and an operation mode selection algorithm is introduced to reduce the access latency. Extensive simulation results show the effects of access-to-update ratio, data popularity, cache size, data object size, and wireless bandwidth. Analytical and simulation results demonstrate that FW-DAS can reduce the access latency with reasonable traffic load compared with poll-each-read (PER)/callback (CB) and their combinations.