Articles in this Volume

Research Article Open Access
Provenance-gated tool invocation for defending LLM agents against indirect prompt injection
Article thumbnail
Indirect prompt injection allows the tool-calling agent to execute calls on untrusted data, and the tool whitelist does not review the source of parameters. This paper introduces provenance gate (PGTI), which sequentially performs tool whitelist and parameter source taint checks before the tool execution, allowing the model to still see all tools. In four AgentDojo tasks, 100 attack pairs, and 60 benign tasks, Qwen2.5 14B ran 960 times. The attack success rate decreased from 17.0% with no defense to 3.0%, and it still held after switching models and attack templates; the utility under attack decreased from 40.0% to 28.0%. The benign utility relative to static tool filtering (35.0% and 26.7%) and the increment over the whitelist alone showed no significant difference in this sample. Pair comparison simultaneously reported both safety and utility.
Show more
Read Article PDF
Cite
Research Article Open Access
Structural design and applications of bio-inspired soft grippers: a review
As robotic systems move beyond structured industrial settings into less structured domains such as agricultural harvesting, food handling, medical manipulation, and human-robot interaction, conventional rigid grippers face limitations when handling soft, fragile, irregularly shaped, or highly variable objects. Soft grippers address this problem by combining compliant materials, deformable structures, and soft actuation to conform to object geometry. Biological systems, including the human hand, bird feet, octopus arms, plant tendrils, and fish fins, offer additional design principles for morphology and motion. This review examines structural design, actuation and materials, stiffness regulation, and representative applications of bio-inspired soft grippers. Gripper architectures are first discussed in terms of discontinuous bending, continuous bending, and continuous twisting or wrapping. Pneumatic, tendon-driven, and smart-material actuation are then reviewed together with additive manufacturing and variable-stiffness techniques. Applications in flexible industrial handling, agricultural harvesting, medical manipulation, and special environments are subsequently considered. The literature indicates that the field is moving beyond the simple pursuit of greater compliance toward coordinated optimization of compliance, load-bearing capability, grasp stability, and control accuracy. Future work is likely to depend increasingly on multimodal sensing, closed-loop control, integrated structure-actuation-sensing design, and intelligent adaptive grasping of previously unseen objects.
Show more
Read Article PDF
Cite
Research Article Open Access
Cooling tower frequency optimization and energy saving potential in a tobacco factory chiller plant with a fixed equipment combination
Article thumbnail
Tobacco manufacturing requires a continuous and stable cooling source for process air conditioning. Chiller-plant optimization must reduce refrigeration and heat-rejection electricity while preserving the established cooling service. In operating archives from industrial plants, however, tower frequency, condenser-water temperature and existing feedback actions often change together, so historical association alone does not identify the temperature response to a new frequency. This study addresses that problem for a tobacco factory chiller plant in Yangzhou by formulating a component energy balance for a fixed equipment combination and treating the unmeasured tower-frequency response as an explicit scenario parameter. The archive ending at 00:00 on 15 October 2026 yields 1372 disjoint windows; the primary combination provides 314 windows over 43 operating dates, divided into 279 fitting, 15 validation and 20 evaluation windows. With measured condenser inlet temperature, evaluation RMSE values are 11.24 kW for chiller power and 0.89 kW for tower power. Reducing three towers from 50 to 47.5 Hz gives a modeled fan-power credit of 4.14 kW; at loads of 2500–4500 kW, the corresponding break-even warming is 0.130–0.072 °C. Under four specified thermal-response scenarios, local optimization gives mean modeled credits of 0.55–10.50 kW, or 0.10%–1.95% of the accounting boundary, and the preferred direction changes with response intensity. The study quantifies the energy tradeoff and the temperature-response measurements required for a subsequent frequency-adjustment test; the scenario credits are not field-measured savings.
Show more
Read Article PDF
Cite
Research Article Open Access
Efficient roadside billboard detection in complex traffic scenes using EBDF-Net
Article thumbnail
Roadside billboards are common visual objects in intelligent traffic, urban monitoring and smart city applications. However, in actual traffic scenarios, scale differences, local occlusion, complex backgrounds and lighting changes can affect detection accuracy and stability. To address these challenges, this paper proposes an Efficient Billboard Detection Framework (EBDF-Net), which combines multi-scale feature refinement and adaptive localisation optimisation to improve feature representation for small and partially obscured billboards while maintaining efficiency in real-time object detection. The framework is specifically designed for roadside billboard detection. By adjusting training parameters and localisation optimisation strategies, it improves the stability and efficiency of bounding-box regression. In addition, data cleaning, annotation optimisation and targeted enhancement are used to improve dataset quality and enhance model robustness under different traffic conditions. The experimental results show that EBDF-Net improves detection performance while maintaining fast inference, making it suitable for traffic scene analysis, roadside facility monitoring and urban monitoring.
Show more
Read Article PDF
Cite
Research Article Open Access
Conformal selective answering with hidden-state probes controls FDR on HaluEval but fails under domain shift in small language models
Article thumbnail
Whether the hidden state probe can provide a false discovery rate (FDR) guarantee for the answer release of the small language model still lacks a cross-domain shift test. This article uses a multi-layer probe to identify the hallucinated answer, decides to release it with conformal selection, and tests it on HaluEval and TruthfulQA. The probe AUROC (area under the ROC curve) on HaluEval is 0.991, which is higher than 0.974 of the answer-length baseline; the FDR at the target level of 0.10 is 0.050, and the power is 0.992, which is higher than the baseline of 0.883. The exploratory AUROC with a single feature of punctuation at the end of the sentence reached 0.969, and the difference in the length and sentence style of HaluEval's answer contributed most of the AUROC. When cross-domain, AUROC dropped to 0.491, and FDR rose to 0.409, exceeding the target level; after recalibration, 0 answers were released. Exploratory in-domain retraining brings AUROC back to 0.904, FDR is 0.045, and power is 0.502. High AUROC cannot be directly interpreted as the model grasping the true and false information, and the probe must be retrained with the data of the target domain.
Show more
Read Article PDF
Cite
Research Article Open Access
Poisoning detection in coded federated learning: impact of linear mixed aggregation on an FLPD-style detector
Article thumbnail
Federated learning (FL) lets clients train a shared model without sharing raw data, but malicious clients can poison it with forged updates whose validity the central server cannot directly verify. Similarity-based detection compares client updates, while coded aggregation reduces uplink cost by mixing updates into packets, whether detection remains effective after such mixing is unclear. This paper evaluates the question on Fashion-MNIST with 100 clients, 10 of them malicious, under two coded aggregation modes and three attacks, namely label flipping, scaled sign reversal and a LIE-style attack. The first two attacks cover mixing sizes m = 1, 2, 5 and 10, the LIE-style variant covers decisive configurations. The LIE-style attack is a purpose-built variant and differs from the published algorithm. In Mode A, decoding precedes detection, which remains lossless in the tested configurations. In Mode B, detection acts on mixed packets and depends on the grouping layout. Under the LIE-style attack at m = 10, individual-level detection coverage is 100%, whereas packet-level coverage is 23% under spread and 100% under consecutive, these units differ, and spread accuracy remains at 86.03%. Under scaled sign reversal at m = 10 with spread, the defense can make the model worse than no defense. Saved traffic can therefore cost detection capability, with outcomes depending on layout as well as mixing size.
Show more
Read Article PDF
Cite