Monocular 3D object detection is a cost-effective solution for applications like autonomous driving and robotics, but remains fundamentally ill-posed due to inherently ambiguous depth cues. Recent DETR-based methods attempt to mitigate this through global attention and auxiliary depth prediction, yet they still struggle with inaccurate depth estimates. Moreover, these methods often overlook instance-level detection difficulty, such as occlusion, distance, and truncation, leading to suboptimal detection performance. We propose MonoDLGD, a novel Difficulty-Aware Label-Guided Denoising framework that adaptively perturbs and reconstructs ground-truth labels based on detection uncertainty. Specifically, MonoDLGD applies stronger perturbations to easier instances and weaker ones into harder cases, and then reconstructs them to effectively provide explicit geometric supervision. By jointly optimizing label reconstruction and 3D object detection, MonoDLGD encourages geometry-aware representation learning and improves robustness to varying levels of object complexity. Extensive experiments on the KITTI benchmark demonstrate that MonoDLGD achieves state-of-the-art performance across all difficulty levels.

Modern HBM-based memory systems have evolved over generations while retaining cache line granularity accesses. Preserving this fine granularity necessitated the introduction of bank groups and pseudo channels. These structures expand timing parameters and control overhead, significantly increasing memory controller scheduling complexity. Large language models (LLMs) now dominate deep learning workloads, streaming contiguous data blocks ranging from several kilobytes to megabytes per operation. In a conventional HBM-based memory system, these transfers are fragmented into hundreds of 32B cache line transactions. This forces the memory controller to employ unnecessarily intricate scheduling, leading to growing inefficiency. To address this problem, we propose RoMe. RoMe accesses DRAM at row granularity and removes columns, bank groups, and pseudo channels from the memory interface. This design simplifies memory scheduling, thereby requiring fewer pins per channel. The freed pins are aggregated to form additional channels, increasing overall bandwidth by$\text{1 2. 5 \%}$with minimal extra pins. RoMe demonstrates how memory scheduling logic can be significantly simplified for representative LLM workloads, and presents an alternative approach for next-generation HBM-based memory systems achieving increased bandwidth with minimal hardware overhead.

A decade after its academic introduction, RowHammer (RH) remains a moving target that continues to challenge both the industry and academia. With its potential to serve as a critical attack vector, the ever-decreasing RH threshold now threatens DRAM process technology scaling, with a superlinearly increasing cost of RH protection solutions. Due to their generality and relatively lower performance costs, architectural RH solutions are the first line of defense against RH. However, the field is fragmented with varying views of the problem, terminologies, and even threat models. In this paper, we systematize architectural RH defenses from the last decade through the lens of streaming algorithms. We provide a taxonomy that encompasses 48 different works. We map multiple architectural RH defenses to the classical streaming algorithms, which extends to multiple proposals that did not identify this link. We also provide two practitioner guides. The first guide analyzes which algorithm best fits a given RHTH, location, process technology, storage type, and mitigative action. The second guide encourages future research to consult existing algorithms when architecting RH defenses. We illustrate this by demonstrating how Reservoir-Sampling can improve related RH defenses, and also introduce StickySampling that can provide mathematical security that related studies do not guarantee.

Transfer learning paradigm has driven substantial advancements in various vision tasks. However, as state-of-the-art models continue to grow, classical full fine-tuning often becomes computationally impractical, particularly in multi-task learning (MTL) setup where training complexity increases proportional to the number of tasks. Consequently, recent studies have explored Parameter-Efficient Fine-Tuning (PEFT) for MTL architectures. Despite some progress, these approaches still exhibit limitations in capturing fine-grained, task-specific features that are crucial to MTL. In this paper, we introduce Task-Adaptive Dynamic transFormer, termed TADFormer, a novel PEFT framework that performs task-aware feature adaptation in the fine-grained manner by dynamically considering task-specific input contexts. TADFormer proposes the parameter-efficient prompting for task adaptation and the Dynamic Task Filter (DTF) to capture task information conditioned on input contexts. Experiments on the PASCAL-Context benchmark demonstrate that the proposed method achieves higher accuracy in dense scene understanding tasks, while reducing the number of trainable parameters by up to 8.4 times when compared to full fine-tuning of MTL models. TADFormer also demonstrates superior parameter efficiency and accuracy compared to recent PEFT methods.

The demand for precise information on DRAM microarchitectures and error characteristics has surged, driven by the need to explore processing in memory, enhance reliability, and mitigate security vulnerability. Nonetheless, DRAM manufacturers have disclosed only a limited amount of information, making it difficult to find specific information on their DRAM microarchitectures. This paper addresses this gap by presenting more rigorous findings on the microarchitectures of commodity DRAM chips and their impacts on the characteristics of activate-induced bitflips (AIBs), such as RowHammer and RowPress. The previous studies have also attempted to understand the DRAM microarchitectures and associated behaviors, but we have found some of their results to be misled by inaccurate address mapping and internal data swizzling, or lack of a deeper understanding of the modern DRAM cell structure. For accurate and efficient reverse-engineering, we use three tools: AIBs, retention time test, and RowCopy, which can be cross-validated. With these three tools, we first take a macroscopic view of modern DRAM chips to uncover the size, structure, and operation of their subarrays, memory array tiles (MATs), and rows. Then, we analyze AIB characteristics based on the microscopic view of the DRAM microarchitecture, such as 6F2cell layout, through which we rectify misunderstandings regarding AIBs and discover a new data pattern that accelerates AIBs. Lastly, based on our findings at both macroscopic and microscopic levels, we identify previously unknown AIB vulnerabilities and propose a simple yet effective protection solution.

Although robot manipulators are widely used in advanced industrial applications, their dynamics has very high complexity and uncertainties, making exact mathematical modeling difficult and preventing high-precision tracking control. Herein, we propose a practical high-performance model-free controller for robot manipulators that attenuates the undesirable side effects of a time-delayed state-based estimation (TDE) technique in terms of the receding-horizon$\mathcal {H}_\infty$performance. By constructing tracking error dynamics with a sliding variable, the effects of TDE errors are identified and suppressed each time in terms of the fixed-horizon$\mathcal {H}_\infty$performance. All initial states are shown to converge to a certain bounded set within a precomputed finite time. The proposed approach is beneficial for sudden transient responses and the temporarily unbounded TDE errors that cannot be handled by existing TDE-based controllers. Finally, the stability of the proposed controller is proven based on the comparison Lemma, and simulations and experiments show that its tracking performance and robustness are superior to those of conventional control algorithms.

We propose an iterative sparse Bayesian policy optimization (ISBPO) scheme as an efficient multitask reinforcement learning (RL) method for industrial control applications that require both high performance and cost-effective implementation. Under continual learning scenarios in which multiple control tasks are sequentially learned, the proposed ISBPO scheme preserves the previously learned knowledge without performance loss (PL), enables efficient resource use, and improves the sample efficiency of learning new tasks. Specifically, the proposed ISBPO scheme continually adds new tasks to a single policy neural network while completely preserving the control performance of previously learned tasks through an iterative pruning method. To create a free-weight space for adding new tasks, each task is learned through a pruning-aware policy optimization method called the sparse Bayesian policy optimization (SBPO), which ensures efficient allocation of limited policy network resources for multiple tasks. Furthermore, the weights allocated to the previous tasks are shared and reused in new task learning, thereby improving sample efficiency and the performance of new task learning. Simulations and practical experiments demonstrate that the proposed ISBPO scheme is highly suitable for sequentially learning multiple tasks in terms of performance conservation, efficient resource use, and sample efficiency.

This paper proposes a new adaptive law based on sliding-mode control(SMC) control. The proposed adaptive law adjusts the switching gain near the sliding manifold to have the appropriate attractivity property. This appropriate attractivity property prevents unexpected errors caused by the excessive or insufficient adaptation of switching gain. Furthermore, this adaptive law ensures the fast adaptation speed of switching gain, and this yields the reduction of chattering. For the more practical implementation based on a model-free controller, the proposed adaptive SMC (ASMC) works together with a time-delay controller(TDC). To validate the proposed adaptive law, the results of simulations are carried out and comparisons are made with existing ASMC control schemes.

A real-time solution for estimating and tracking the 3D pose of a rigid object is presented for image-based visual servo with natural landmarks. The many state-of-the-art technologies that are available for recognizing the 3D pose of an object in a natural setting are not suitable for real-time servo due to their time lags. This paper demonstrates that a real-time solution of 3D pose estimation become feasible by combining a fast tracker such as KLT [7] [8] with a method of determining the 3D coordinates of tracking points on an object at the time of SIFT based tracking point initiation, assuming that a 3D geometric model with SIFT description of an object is known a-priori. Keeping track of tracking points with KLT, removing the tracking point outliers automatically, and reinitiating the tracking points using SIFT once deteriorated, the 3D pose of an object can be estimated and tracked in real-time. This method can be applied to both mono and stereo camera based 3D pose estimation and tracking. The former guarantees higher frame rates with about 1 ms of local pose estimation, while the latter assures of more precise pose results but with about 16 ms of local pose estimation. The experimental investigations have shown the effectiveness of the proposed approach with real-time performance.

An approach to identifying a caller or callers by a service robot is presented for a natural interaction with people in a home/office environment. The problem addressed specifically in this paper is how to successfully identify a caller in a cluttered environment with large uncertainties involved in the sensed audio-visual cues. The proposed approach is based on a proposition that the dependability of perceptual recognition may come unlikely from "the effort to make individual sensing perfect", but likely from "the effort to self-generate perceptual behaviors of integrating individual sensing that lead to mission accomplishment, no matter how imperfect and uncertain individual sensing may be". We implement the above proposition in terms of a novel robotic architecture, referred to here as "cognitive robotic engine (CRE)." CRE implemented for the case of a robot identifying a caller in a crowded and noisy environment, including its experimental results, are shown

There has shown a significant interest in a high performance of, at the same time, a compact size and low cost of, 3D sensor, in reflection of a growing need of 3D environmental sensing for service robotics. One of the important requirements associated with such a 3D sensor is that sensing does not irritate or disturb human in any way while working in close and continuous contact with human. Furthermore, such a 3D sensor should be reliable and robust to the change of environmental illumination as service robots are required to work day and night. This paper presents a 3D IR camera with variable structured light that is human friendly and robust enough for application to home service robots. Infrared is chosen as the sensing medium in order to meet the requirement of human friendliness and robustness to illumination change. A Digital Mirror Device (DMD) is employed to generate and project variable patterns at a high speed for real-time operation. In implementation, we emphasize the integration of modular components to support real-time sensing and compactness in size. A number of real-world experimentations are conducted, including a human face, a statue, and a plastic model. The experimental results have demonstrated that the implemented 3D IR Camera is robust to illumination change, in addition to its advantage of human friendliness.