Developing robotic hands that adapt to real-world dynamics remains a fundamental challenge in robotics and machine intelligence. Despite notable advances in replicating human-hand kinematics and control algorithms, robotic systems still struggle to match human capabilities in dynamic environments, primarily due to inadequate tactile feedback. To bridge this gap, we present F-TAC Hand, a biomimetic hand featuring high-resolution tactile sensing (0.1-mm spatial resolution) across 70% of its surface area. Through optimized hand design, we overcome traditional challenges in integrating high-resolution tactile sensors while preserving the full range of motion. The hand, powered by our generative algorithm that synthesizes human-like hand configurations, demonstrates robust grasping capabilities in dynamic real-world conditions. Extensive evaluation across 600 real-world trials demonstrates that this tactile-embodied system significantly outperforms non-tactile-informed alternatives in complex manipulation tasks (P < 0.0001). These results provide empirical evidence for the critical role of rich tactile embodiment in developing advanced robotic intelligence, offering promising perspectives on the relationship between physical sensing capabilities and intelligent behaviour.

Real-time recognition of locomotion-related activities is a fundamental skill that a controller of lower limb wearable robots should possess. Subject-specific training and reliance on electromyographic interfaces are the main limitations of existing approaches. This study presents a novel methodology for real-time locomotion mode recognition of locomotion-related activities in lower limb wearable robotics. A hybrid classifier can distinguish among seven locomotion-related activities. First, a time-based approach classifies between static and dynamical states based on gait kinematics data. Second, an event-based fuzzy-logic method triggered by foot pressure sensors operates in a subject-independent fashion on a minimal set of relevant biomechanical features to classify among dynamical modes. The locomotion mode recognition algorithm is implemented on the controller of a portable powered orthosis for hip assistance. An experimental protocol is designed to evaluate the controller performance in an out-of-lab scenario without the need for subject-specific training. Experiments are conducted on six healthy volunteers performing locomotion-related activities at slow, normal, and fast speeds under the zero-torque and assistive mode of the orthosis. The overall accuracy rate of the controller is 99.4% over more than 10 000 steps, including seamless transitions between different modes. The experimental results show a successful subject-independent performance of the controller for wearable robots assisting locomotion-related activities.

This paper presents a novel strategy aiming to acquire an accurate and walking-speed-adaptive estimation of the gait phase through noncontact capacitive sensing and adaptive oscillators (AOs). The capacitive sensing system is designed with two sensing cuffs that can measure the leg muscle shape changes during walking. The system can be dressed above the clothes and free human skin from contacting to electrodes. In order to track the capacitance signals, the gait phase estimator is designed based on the AO dynamic system due to its ability of synchronizing with quasi-periodic signals. After the implementation of the whole system, we first evaluated the offline estimation performance by experiments with 12 healthy subjects walking on a treadmill with changing speeds. The strategy achieved an accurate and consistent gait phase estimation with only one channel of capacitance signal. The average root-mean-square errors in one stride were 0.19 rad (3.0% of one gait cycle) for constant walking speeds and 0.31 rad (4.9% of one gait cycle) for speed transitions even after the subjects rewore the sensing cuffs. We then validated our strategy in a real-time gait phase estimation task with three subjects walking with changing speeds. Our study indicates that the strategy based on capacitive sensing and AOs is a promising alternative for the control of exoskeleton/orthosis.

Human intent recognition is important to the control of robotic prosthesis. In this paper, we propose a multi-level real-time on-board system to recognize continuous locomotion modes. A cascaded classification strategy is designed for the recognition of six steady locomotion modes and 10 transitions. On-board signals of the robotic prosthesis include two inertial measurement units and one load cell. Three transtibial amputees are recruited in the experiments. The prediction decision time of the real-time on-board cascaded classification system is about 3.3 ms, which is enough short compared with the sliding window increment 10 ms. It is easy to recognize the standing and ambulation in the first-level classification with a 99.86% accuracy by quadratic discriminant analysis (QDA) classifier. In the second-level classification, threshold method is adopted to divide one stride into swing and stance phases. In swing phase, five steady modes are recognized with a total accuracy of 96.40%. In stance phase, all these five steady modes are recognized with a total accuracy of 91.21%. The average recognition accuracy of the three subjects is 93.21% by QDA classifier. Besides, for transitions, the proposed system could recognize all transitions rightly. The designed system is feasible and effective to realize real-time on-board recognition of continuous locomotion modes, which is promising for the further control of the prosthesis.

Locomotion intent prediction is essential for the control of powered lower-limb prostheses to realize smooth locomotion transitions. In this research, we develop a multi-sensor fusion based locomotion intent prediction system, which can recognize current locomotion mode and detect locomotion transitions in advance. Seven able-bodied subjects were recruited for this research. Signals from two foot pressure insoles and three inertial measurement units (one on the thigh, one on the shank and the other on the foot) are measured. A two-level recognition strategy is used for the recognition with linear discriminate classifier. Six kinds of locomotion modes and ten kinds of locomotion transitions are tested in this study. Recognition accuracy during steady locomotion periods (i.e., no locomotion transitions) is 99.71% ± 0.05% for seven able-bodied subjects. During locomotion transition periods, all the transitions are correctly detected and most of them can be detected before transiting to new locomotion modes. No significant deterioration in recognition performance is observed in the following five hours after the system is trained, and small number of experiment trials are required to train reliable classifiers.