
Legged robots are increasingly inspired by insects’ ability to combine rigidity and compliance for agile, damage-tolerant locomotion in cluttered terrain. Herein, we introduce INSECTRON, a centimeter-scale hexapod robot that employs magnetic transmission to achieve robust mobility with minimal mechanical complexity. Each leg connects to the drivetrain via neodymium magnets, forming a self-aligning micro-clutch that transmits torque under normal load and safely slips under overload, while additional magnetic couplings between drive modules accommodate load imbalances and perturbations. Hall effect sensors provide proprioception for a phase-domain controller maintaining persistent anti-phase coordination. The magnetic transmission architecture introduces intrinsic mechanical adaptability, functionally analogous to biological preflex mechanisms, enabling rapid responses to external perturbations without additional control complexity. Experiments demonstrated stable locomotion across multiple slopes and substrates. Maneuverability was achieved through differential turning and spin-in-place. A minimum Cost of Transport (CoT) of 30 was achieved during locomotion experiments. Under vertical perturbations, INSECTRON autonomously recovered in over 90% of trials, highlighting strong passive impact tolerance. The architecture eliminates the need for precision alignment and mechanical fasteners, enabling tool-free leg replacement and rapid mechanical reconfiguration. This work establishes a scalable route toward modular, resilient, and field-deployable miniature legged robots for confined and uncertain environments.
Abstract Hand exoskeletons and functional electrical stimulation (FES) have shown promise in restoring hand function after stroke or cervical spinal cord injury. However, major challenges remain in providing reliable user command input in individuals with severe paralysis and in enabling safe and effective control of hand closing and opening. Here, we present a non-invasive tongue-based control strategy for a hybrid soft hand exoskeleton system combining tongue-controlled independent tendon-driven flexion of five fingers with tongue-triggered FES-induced finger extension. This user-intent-driven hybrid approach supports multiple activities of daily living, such as grasping a bottle or toothbrush, and enables hand opening while mitigating muscle fatigue and reducing system complexity. The system was evaluated in five users with cervical spinal cord injury, demonstrating high task success rates in daily living tasks and user preference due to its wearability and ease of use, as well as consistently low reported levels of discomfort, pain, and fatigue. These findings show that the proposed hybrid system can restore activities of daily living in individuals with cervical tetraplegia, including those with long-term injuries.
How wing morphology shapes flight performance is a central question in functional biology and bio-inspired robotics, yet these relationships are difficult to isolate in living organisms. Here, we use a robotic butterfly with tunable wing geometry to systematically characterize morphology–performance relationships by combining motion capture with morphometric analysis. We then use deep reinforcement learning to search the morphological design space and identify wing designs optimized for aerodynamic performance. The resulting designs increase lift and thrust performance by 339% and 46%, respectively. Physical flight tests validate the learning-derived morphology, with prediction errors remaining below 7%. By integrating controlled robotic experiments with machine learning, this framework provides a means to generate testable hypotheses in functional morphology while guiding the engineering design of bio-inspired robots.
Abstract Tactile perception is fundamental to manipulation and sensing, yet marker design in vision-based tactile sensors (VTS) remains largely unexplored. Conventional approaches assume that marker displacement alone suffices to characterise soft tissue deformation, without optimising marker geometry or spatial distribution. We introduce a systematic and design-centric framework based on Marker Information Efficacy (MIE) that formalises how marker properties govern the encoding of deformation information. Starting from generic marker matrices, we systematically examine how marker geometry and spatial distribution affect deformation encoding, force prediction, and computational efficiency. This analysis reveals that marker performance depends on how effectively marker design captures the dominant structure of the deformation field, thereby enabling the principled design of deformation-aligned markers for different sensing scenarios. As a representative case, the derived deformation-aligned marker demonstrates that MIE-guided marker design enables accurate force prediction using simpler predictive models with reduced computational cost. These findings establish MIE as a general and extensible design framework for VTS, effectively transforming optical markers from passive tracking tools into deformation-information-encoding components, and enabling efficient, real-time tactile sensing in contact-rich robotic manipulation.
Robot-assisted minimally invasive surgery lacks haptic feedback for localizing subsurface tumors and estimating depth. We present a compact, high-density fiber Bragg grating (FBG) tactile sensor array for robotic palpation. A hexagonal close-packed topology increases spatial sampling within an 8.5-mm maximum lateral span while maintaining high sensitivity. Static and dynamic tests demonstrate linear force-wavelength responses across seven channels, a worst-case nominal interrogator-limited force resolution of 8.43 mN, and <1% full-scale dynamic error. We further introduce FBG-PatchFormer, which uses contrastive self-supervised pretraining on unlabeled palpation windows followed by supervised finetuning for depth classification. In phantoms with embedded inclusions, the model achieved 99.60% record-level accuracy for ten classes (blank and 2–10 mm) within the calibrated target-stiffness regime. Cross-condition tests showed transfer across inclusion shape and moderate surface curvature; performance under background-stiffness shift was partially recovered using three labeled records per class. In vivo porcine-liver experiments demonstrated force-map-based localization under physiological motion and fluid interference, supporting the feasibility of the sensing system for robotic palpation.