Ying Tan

Assistive robots lack natural proprioceptive feedback, often forcing users to rely on vision. With recent advances, haptic interfaces can now accommodate the multidimensional nature of proprioceptive information. Skin-stretch is one promising haptic modality to provide proprioceptive information that resembles natural body mechanisms and conveys both magnitude and direction information, enabling compact multi-channel designs. While prior studies have examined perceptibility and channel interference, the ability of participants to integrate information across multiple channels remains not well understood. This study evaluates how effectively users can combine two skin-stretch channels to interpret two-dimensional cues. 21 able-bodied participants judged, in collocated and non-collocated placement, the linear projection of stimuli under four patterns of skin-stretch variation. Participants could generally perceive and integrate both channels into a coherent understanding of two-dimensional cues, though some non-uniformity in interpretation was observed. Channel proximity showed no clear effect on performance or workload. These findings provide the first systematic investigation of multi-channel integration in skin-stretch feedback and inform the design of scalable proprioceptive haptic feedback interfaces for assistive robots.

Soft pneumatic actuators (SPAs) are widely used in applications requiring safe and compliant interaction; however, achieving bidirectional motion within a compact and predictable architecture remains a key challenge. Existing approaches typically rely on antagonistic actuator pairs or multi-chamber designs, which increase system complexity and control requirements, while single-chamber solutions often lack robust analytical models to predict their mechanical response. In this work, a Bidirectional Origami-Inspired Soft Pneumatic Actuator (Bi-OSPA) is proposed to achieve both elongation and contraction within a single-chamber structure, where the direction of motion is governed solely by the applied pressure (vacuum or positive). The actuator leverages origami-inspired geometry, allowing deformation to be primarily described through folding kinematics, which facilitates analytical modeling. An analytical framework is developed to predict actuator deformation as well as the corresponding elastic and output forces based on geometric parameters and pressure input, and is validated experimentally, showing good agreement across the displacement range. Furthermore, the effects of key design parameters on displacement and force output are investigated and characterized. The proposed Bi-OSPA combines structural predictive capability and bidirectional functionality, providing a foundation for the design and optimization of soft actuators. Its versatility is further demonstrated through applications in achieving pure twisting when integrated with a Kresling origami unit and as an actuation unit for a one-degree-of-freedom robotic finger enabling flexion and extension.

In daily activities, humans rely on visual information to plan hand movements, making the extraction of task-relevant information through eye gaze a key aspect of motor control. Behavioral studies have revealed characteristic saccade-pursuit patterns, likely governed by shared neural circuits, which enable an efficient reduction of task-related uncertainty. However, a unifying computational principle explaining the emergence of these patterns in continuous tasks such as reading or driving is still lacking. Here we propose a dual stochastic model predictive control formulation of active vision, in which eye movements are continuously controlled to minimize task-relevant uncertainty and build an internal model used for hand movement planning. Through experiments manipulating the amount, density, and difficulty of future visual information, we show how eye movement patterns adapt to the information context while maintaining an invariant extraction horizon. A saccade-pursuit pattern naturally emerges from the model, which accurately predicts both eye and hand movement features observed in experiments. These results provide a principled framework for understanding the continuous regulation of human eye movements and open new perspectives for applications in robotic assistance and active perception.

This paper presents a discrete-time extremum seeking control (ESC) scheme for multi-input single-output (MISO) Hammerstein systems with unknown linear dynamics and a multi-input multi-output (MIMO) cost function. The steady-state cost function may have multiple extrema and, combined with the linear dynamics, exhibits complex behavior; it is assumed to be differentiable, nonconvex, and accessible only via noisy, dynamically filtered sensor measurements. The proposed algorithm integrates stochastic dithers with a stochastic approximation algorithm with expanding truncations (SAAWET), ensuring bounded iterates and closed-loop stability under unmodeled sensor dynamics. Convergence to the set of stationary points of the composite cost function is established from arbitrary initial conditions. The framework extends discrete-time ESC theory to nonconvex, high-dimensional settings with measurement noise and dynamic sensing, providing a rigorous foundation for real-time optimization of nonlinear dynamic systems.

Monitoring forearm muscle activity is essential for estimating human intention in different applications such as virtual reality, rehabilitation, and prosthetics. While prior studies have mainly focused on recognising gestures or simultaneously classifying gestures and force, predominantly using surface electromyography (sEMG), the simultaneous classification of gesture, force, and hand aperture within the same gesture has not yet been investigated. This work extends previous approaches by evaluating a wearable Displacement Myography (DMG)-based system for the combined recognition of hand gestures, four grasping force levels, and multiple hand aperture levels, and by directly benchmarking it against an sEMG system under identical experimental conditions. Data from 13 healthy participants were collected using the MyoLog armband with DMG sensors and sEMG sensors while performing three grasp types (pinch, key, power) with objects of varying size and four force levels, generating 33 configurations. Across several classifiers, the DMG system achieved a peak accuracy of 93 ± 2%, significantly outperforming EMG (64 ± 10%). DMG showed superior discrimination of hand aperture levels (98 ± 3%), while EMG showed relatively better but still lower performance in force-level classification (82 ± 11%). The results are supported by separability analysis using the Fisher Discriminant Ratio, which demonstrated consistently higher class separability for DMG. These findings highlight the suitability of DMG for applications requiring precise recognition of hand aperture and force levels, offering a compact, computationally efficient, and stimulation-compatible solution that overcomes several limitations of EMG-based systems.

Proximity operations of rigid bodies, such as spacecraft rendezvous and docking, require precise tracking of both position and attitude over finite time intervals. These operations are often repeated under uncertain conditions, with unknown but repeatable parameters and disturbances. Adaptive iterative learning control (ILC) is well suited to such tasks, as it can track desired trajectories while learning unknown, iteration-invariant signals or parameters. However, conventional adaptive ILC faces two challenges: (i) the coupling between rotational and translational dynamics complicates the design of the two coordinated learning loops for position and attitude, and (ii) standard adaptive ILC designs cannot guarantee bounded control inputs. To address these issues, we propose a dual-number-based, segment-based two-loop adaptive ILC framework for simultaneous high-precision position and attitude tracking. The framework employs two learning loops that interact through a dual-number representation of tracking errors, combining position and attitude errors into a single mathematical object for unified control design. A segment-based dynamic projection mechanism ensures that both parameter estimates and control inputs remain bounded without prior knowledge of uncertainties. Mathematical analysis and numerical simulations demonstrate that the proposed framework significantly enhances tracking performance under unknown but repeatable uncertainties and strong rotational-translational coupling.

Fine-tuning large language models (LLMs) using standard first-order (FO) optimization often drives training toward sharp, poorly generalizing minima. Conversely, zeroth-order (ZO) methods offer stronger exploratory behavior without relying on explicit gradients, yet suffer from slow convergence. More critically, our analysis reveals that in generative tasks, the vast output and search space significantly amplify estimation variance, rendering ZO methods both noisy and inefficient. To address these challenges, we propose \textbf{Hi-ZFO} (\textbf{Hi}erarchical \textbf{Z}eroth- and \textbf{F}irst-\textbf{O}rder optimization), a hybrid framework designed to synergize the precision of FO gradients with the exploratory capability of ZO estimation. Hi-ZFO adaptively partitions the model through layer-wise importance profiling, applying precise FO updates to critical layers while leveraging ZO optimization for less sensitive ones. Notably, ZO in Hi-ZFO is not merely a memory-saving surrogate; it is intentionally introduced as a source of "beneficial stochasticity" to help the model escape the local minima where pure FO optimization tends to stagnate. Validated across diverse generative, mathematical, and code reasoning tasks, Hi-ZFO consistently achieves superior performance while significantly reducing the training time. These results demonstrate the effectiveness of hierarchical hybrid optimization for LLM fine-tuning.