Yuquan Wang

Kai Pfeiffer, Quan Zhang, Yuqing Chen et al.

This work presents a novel and efficient nonlinear programming framework that tightly integrates hierarchical decision-making with whole-body inverse kinematic planning and control. Decision-making plays a central role in many aspects of robotics, from sparse inverse kinematic control with a minimal number of joints, to inverse kinematic planning while simultaneously selecting a discrete end-effector location from multiple candidates. Current approaches often rely on heavy computations using mixed-integer nonlinear programming, separate decision-making from inverse kinematics (some times approximated by reachability methods), or employ efficient but less versatile $\ell_1$-norm formulations of linear sparse programming, without addressing the underlying nonlinear problem formulations. In contrast, the proposed sparse hierarchical nonlinear programming solver is efficient, versatile, and accurate by exploiting sparse hierarchical structure and leveraging the $\ell_0$-norm which is rarely used in robotics. The solver efficiently tackles complex nonlinear hierarchical decision-making problems previously unaddressed in the literature, such as inverse kinematic planning with simultaneous prioritized selection of end-effector locations from a large set of candidates, or inverse kinematic control with simultaneous selection of bi-manual grasp locations on a randomly rotated box.

Qingyang Zhong, Jifan Yu, Zheyuan Zhang et al.

Adaptive learning aims to stimulate and meet the needs of individual learners, which requires sophisticated system-level coordination of diverse tasks, including modeling learning resources, estimating student states, and making personalized recommendations. Existing deep learning methods have achieved great success over statistical models; however, they still lack generalization for diverse tasks and suffer from insufficient capacity since they are composed of highly-coupled task-specific architectures and rely on small-scale, coarse-grained recommendation scenarios. To realize the idea of general adaptive systems proposed in pedagogical theory, with the emerging pre-training techniques in NLP, we try to conduct a practical exploration on applying pre-training to adaptive learning, to propose a unified framework based on data observation and learning style analysis, properly leveraging heterogeneous learning elements. Through a series of downstream tasks of Learning Recommendation, Learning Resource Evaluation, Knowledge Tracing, and Dropout Prediction, we find that course structures, text, and knowledge are helpful for modeling and inherently coherent to student non-sequential learning behaviors and that indirectly relevant information included in the pre-training foundation can be shared across downstream tasks to facilitate effectiveness. We finally build a simplified systematic application of adaptive learning and reflect on the insights brought back to pedagogy. The source code and dataset will be released.

Xingzhou Chen, Xiayan Xu, Yan Ning et al.

Humanoid robots are entering our physical world at scale, yet as oversized toys--good at singing and dancing, but short on force-interaction capabilities for practical tasks. Bridging this gap necessitates prioritizing reliable contact perception as a fundamental requirement. Estimating external wrenches in humanoids is complicated by floating-base dynamics and indeterminate contact locations. Existing analytical frameworks require idealistic assumptions and hard-to-obtain measurements, which are often unavailable in practice. To bridge this gap, we propose SixthSense, a task-agnostic approach that infers whole-body contact timing, location, and wrenches from proprioception and IMU data alone. To capture the multi-modal dynamics between unstructured contact inputs and the uncertain motion outputs, we employ conditional flow matching to tokenize proprioceptive histories and estimate a spatiotemporally sparse contact-event flow. SixthSense serves as a plug-and-play perception module for applications including collision detection, physical human-robot interaction, and force-feedback teleoperation. Experiments across standing, walking, and whole-body motion-tracking policies showcased unprecedented performance in diverse behaviors.

Xiaoyang Liu, Yuquan Wang, Zheng Chen et al.

Currently, methods for single-image deblurring based on CNNs and transformers have demonstrated promising performance. However, these methods often suffer from perceptual limitations, poor generalization ability, and struggle with heavy or complex blur. While diffusion-based methods can partially address these shortcomings, their multi-step denoising process limits their practical usage. In this paper, we conduct an in-depth exploration of diffusion models in deblurring and propose a one-step diffusion model for deblurring (OSDD), a novel framework that reduces the denoising process to a single step, significantly improving inference efficiency while maintaining high fidelity. To tackle fidelity loss in diffusion models, we introduce an enhanced variational autoencoder (eVAE), which improves structural restoration. Additionally, we construct a high-quality synthetic deblurring dataset to mitigate perceptual collapse and design a dynamic dual-adapter (DDA) to enhance perceptual quality while preserving fidelity. Extensive experiments demonstrate that our method achieves strong performance on both full and no-reference metrics. Our code and pre-trained model will be publicly available at https://github.com/xyLiu339/OSDD.

Yining Wang, Yuquan Wang, Xi Li et al.

As Large Language Models (LLMs) continue to advance, they are increasingly relied upon as real-time sources of information by non-expert users. To ensure the factuality of the information they provide, much research has focused on mitigating hallucinations in LLM responses, but only in the context of formal user queries, rather than maliciously crafted ones. In this study, we introduce The Illusionist's Prompt, a novel hallucination attack that incorporates linguistic nuances into adversarial queries, challenging the factual accuracy of LLMs against five types of fact-enhancing strategies. Our attack automatically generates highly transferrable illusory prompts to induce internal factual errors, all while preserving user intent and semantics. Extensive experiments confirm the effectiveness of our attack in compromising black-box LLMs, including commercial APIs like GPT-4o and Gemini-2.0, even with various defensive mechanisms.

Yuquan Wang, Mi Zhang, Yining Wang et al.

Large Reasoning Models (LRMs) have demonstrated impressive performance in reasoning-intensive tasks, but they remain vulnerable to harmful content generation, particularly in the mid-to-late steps of their reasoning processes. Existing defense mechanisms, however, rely on costly fine-tuning and additional expert knowledge, which restricts their scalability. In this work, we propose ReasoningGuard, an inference-time safeguard for LRMs, which injects timely safety aha moments to steer harmless while helpful reasoning processes. Leveraging the model's internal attention behavior, our approach accurately identifies critical points in the reasoning path, and triggers spontaneous, safety-oriented reflection. To safeguard both the subsequent reasoning steps and the final answers, we further implement a scaling sampling strategy during the decoding phase, selecting the optimal reasoning path. Inducing minimal extra inference cost, ReasoningGuard effectively mitigates three types of jailbreak attacks, including the latest ones targeting the reasoning process of LRMs. Our approach outperforms seven existing safeguards, achieving state-of-the-art safety defenses while effectively avoiding the common exaggerated safety issues.

Yuquan Wang, Niels Dehio, Abderrahmane Kheddar

In order to enable on-purpose robotic impact tasks, predicting joint-velocity jumps is essential to enforce controller feasibility and hardware integrity. We observe a considerable prediction error of a commonly-used approach in robotics compared against 250 benchmark experiments with the Panda manipulator. We reduce the average prediction error by 81.98% as follows: First, we focus on task-space equations without inverting the ill-conditioned joint-space inertia matrix. Second, before the impact event, we compute the equivalent inertial properties of the end-effector tip considering that a high-gains (stiff) kinematic-controlled manipulator behaves like a composite-rigid body.

Yuquan Wang, Niels Dehio, Abderrahmane Kheddar

State-of-the-art impact dynamics models either apply for free-flying objects or do not account that a robotic manipulator is commonly high-stiffness controlled. Thus, we lack tailor-made models for manipulators mounted on a fixed base. Focusing on orthogonal point-to-surface impacts (no tangential velocities), we revisit two main elements of an impact dynamics model: the contact-force model and the inverse inertia matrix. We collect contact-force measurements by impacting a 7 DOF Panda robot against a sensorized rigid environment with various joint configurations and velocities. Evaluating the measurements from 150 trials, the best model-to-data matching suggests a viscoelastic contact-force model and computing the inverse inertia matrix assuming the robot is a composite-rigid body.

Yuquan Wang, Niels Dehio, Arnaud Tanguy et al.

Robots usually establish contacts at rigid surfaces with near-zero relative velocities. Otherwise, impact-induced energy propagates in the robot's linkage and may cause irreversible damage to the hardware. Moreover, abrupt changes in task-space contact velocity and peak impact forces also result in abrupt changes in robot joint velocities and torques; which can compromise controllers' stability, especially for those based on smooth models. In reality, several tasks would require establishing contact with moderately high velocity. We propose to enhance task-space multi-objective controllers formulated as a quadratic program to be resilient to frictional impacts in three dimensions. We devise new constraints and reformulate the usual ones to be robust to the abrupt joint state changes mentioned earlier. The impact event becomes a controlled process once the optimal control search space is aware of: (1) the hardware-affordable impact bounds and (2) analytically-computed feasible set (polyhedra) that constrain post-impact critical states. Prior to and nearby the targeted contact spot, we assume, at each control cycle, that the impact will occur at the next iteration. This somewhat one-step preview makes our controller robust to impact time and location. To assess our approach, we experimented its resilience to moderate impacts with the Panda manipulator and achieved swift grabbing tasks with the HRP-4 humanoid robot.

Yuquan Wang, Arnaud Tanguy, Pierre Gergondet et al.

Impact-aware tasks (i.e. on purpose impacts) are not handled in multi-objective whole body controllers of hu-manoid robots. This leads to the fact that a humanoid robot typically operates at near-zero velocity to interact with the external environment. We explicitly investigate the propagation of the impact-induced velocity and torque jumps along the structure linkage and propose a set of constraints that always satisfy the hardware limits, sustain already established contacts and the stability measure, i.e. the zero moment point condition. Without assumptions on the impact location or timing, our proposed controller enables humanoid robots to generate non-zero contact velocity without breaking the established contacts or falling. The novelty of our approach lies in building on existing continuous dynamics whole body multi-objective controller without the need of reset-maps or hybrid control.