Exoskeleton Robot Human Intention Recognition 2026
Exoskeleton Human Intention Recognition Patents
Human intention recognition (HIR) is the core capability determining whether an exoskeleton acts as a biomechanical extension or a poorly-synchronized device. This dataset spans 2012–2026 across biosignal, kinematic, and multi-modal sensing approaches.
How Exoskeletons Decode Wearer Intent
Human intention recognition for exoskeleton robots encompasses the sensing, signal processing, and classification pipeline that translates physiological and kinematic cues into actionable control commands. Within this dataset, four core signal domains are represented: biosignal-based sensing (sEMG and EEG), inertial and kinematic sensing (IMUs, encoders, force sensors), novel contactless sensing (Hall effect sensors, magnetic fields, camera vision), and multi-modal fusion combining two or more modalities.
Surface electromyography remains the most clinically validated modality for pre-movement intention capture, encoding motor intent before mechanical movement is observable. Deep learning models — particularly LSTMs and recurrent neural networks — map sEMG signal patterns to joint angles or motion classes. A key limitation is high inter-subject variability requiring individual calibration, which constrains plug-and-play deployment as noted in a 2019 study on EMG-based movement onset detection for upper-limb exoskeletons.
IMU arrays, rotary encoders, force sensors, and pressure insoles provide body kinematics without the noise susceptibility of biosignals, making this modality preferred for industrial and assistive applications. A 2022 study achieved 97.35% accuracy across 8 activity classes with 0.506-second end-to-end latency using five deep learning models deployed on an edge device, establishing a concrete performance benchmark for real-time HIR in wearable systems.
Multi-modal fusion combining sEMG with IMUs, force sensors, or encoders is identified in this dataset as the dominant performance-improvement strategy, with a 2022 BiLSTM-based rehabilitation robot study reaching 99.61% recognition accuracy. Innovation in retrieved records is distributed across many players rather than concentrated, suggesting a field in competitive expansion rather than consolidation.
Filing Activity and Performance Benchmarks
The retrieved dataset spans 2012–2026 and shows a clear three-phase trajectory from foundational sensor architectures to edge-deployed deep learning systems. Performance benchmarks and filing dates reflect innovation signals within this dataset only.
HIR Accuracy Benchmarks by Approach (Retrieved Records)
In this dataset, multi-modal BiLSTM fusion achieves the highest reported accuracy at 99.61%, followed by IMU deep learning at 97.35%, with EMG-encoder fusion at 92% and single-modality EMG at 82.1%.
↗ Click bars to explorePatent and Literature Publication Activity by Phase (Dataset Snapshot)
In this dataset, publication activity accelerates markedly in the Maturation phase (2022–2026), reflecting growing edge-AI and vision-based HIR filings compared to the Foundational phase (2012–2017).
↗ Click bars to exploreWhere HIR Technology Is Deployed Across Sectors
Within this dataset, HIR for exoskeleton robots is applied across five sectors: medical rehabilitation, industrial ergonomics, defense, teleoperation, and prosthetics. Each sector drives distinct sensing and algorithmic requirements.
Lower-Limb Rehabilitation Exoskeletons
Medical rehabilitation is the largest application sector in this dataset, targeting stroke patients, paraplegics, and the elderly. A 2021 study applied IPSO-LSTM to map sEMG signals to continuous lower-limb joint angles, while a 2022 BiLSTM multi-source fusion approach achieved 99.61% recognition accuracy for a rehabilitation robot. A 2021 review covered control strategies for lower-limb rehabilitation exoskeletons across gait phase detection and balance state estimation.
Medical RehabilitationIndustrial Ergonomics Worker Assistance
Industrial exoskeletons deployed in manufacturing and logistics require robust HIR across unscripted motion sequences. A 2021 study on soft-robotic exoskeletons used IMU-based action recognition for battery-life optimization through action-dependent control. A separate 2021 study introduced a hybrid human-exoskeleton digital model for digital twin monitoring of physiological status in assembly and logistics workplaces.
Industrial ErgonomicsDefense and Physical Augmentation
Military-funded research underlies several key patents in this dataset. The Enhanced Activated Exoskeleton System (APTIMA, Inc., US, 2023) was developed under US Army contract W911NF-17-C-0062 and uses machine recognition of user intention for exoskeleton control. The Intelligent Robotic Framework patent (Mahendra Engineering College, IN, 2026) explicitly lists defense operations as an application domain, combining autonomous humanoid robots with human-assistive exoskeleton systems.
DefenseTeleoperation and Remote Manipulation
Upper-limb exoskeletons used as motion-capture interfaces for controlling remote robots depend on accurate joint-angle estimation and low-latency HIR. A 2023 study presented a passive, lightweight wearable upper-limb exoskeleton with upper-limb pose estimation enabling 14-DOF remote anthropomorphic manipulator control. Locomotion intent recognition for lower-limb prostheses — reviewed in a 2021 paper — shares methodology with exoskeleton HIR across locomotion mode recognition, gait event detection, and continuous gait phase estimation.
TeleoperationKey Patent Assignees in Exoskeleton HIR — Dataset Snapshot
In this dataset, nine named patent assignees are identifiable across US, CN, KR, CA, WO, and IN jurisdictions. No single assignee in retrieved records dominates the field; innovation is distributed across commercial firms, research universities, and government-funded entities, reflecting a field in competitive expansion.
Top Assignees by Filing Activity in Retrieved Records (Dataset Snapshot)
↗ Click bars to exploreUniversity of Science and Technology of China
The University of Science and Technology of China (and its Institute of Advanced Technology) holds two CN patents in this dataset, spanning 2020–2022. The 2020 patent covers a neural network control system integrating facial feature capture, sEMG, and gyroscopes into an intent classification pipeline. The 2022 active patent introduces an intent assimilation control method for upper-limb exoskeletons using sEMG, covering a continuous interaction spectrum from cooperative to competitive human-robot modes.
China — CNHyundai Motor Company
Hyundai Motor Company filed a US patent in 2024 for a perception system for a lower-body powered exoskeleton, integrating camera-based terrain sensing with augmented reality (AR) or mixed reality (MR) visualization to relay terrain feedback to the wearer in real time. This is one of only two dataset entries addressing camera-based footstep planning and AR user feedback for powered lower-body exoskeletons, alongside a parallel 2024 WO filing from Boston Dynamics. The patent signals Hyundai’s focus on vision-augmented intention support for community ambulatory devices.
South Korea — KRFive Directional Signals in HIR (2023–2026)
Based on the most recent filings and publications in this dataset (2023–2026), five directional signals indicate where HIR technology is heading, from anticipatory synchronization to AI-integrated multi-domain robotic frameworks.
Anticipatory and Predictive Synchronization
The MIT WO 2026 patent on real-time anticipation and synchronization in close-proximity human-robot collaboration represents a paradigm shift from classifying current intent to predicting future intent before movement occurs, enabling anticipatory robot response rather than corrective response. This moves beyond reactive recognition architectures and may render classification-then-response pipelines obsolete for high-performance applications. IP strategists should evaluate whether their control systems could be affected by the MIT WO patent claim scope.
Vision and AR Terrain-Adaptive Control
Hyundai Motor Company (US, 2024) and Boston Dynamics (WO, 2024) both filed patents addressing camera-based terrain mapping combined with AR user feedback as an intention-augmentation layer, signaling convergence of computer vision with bidirectional environmental understanding. These are the only two dataset entries addressing this approach for powered lower-body exoskeletons. The domain is described as relatively uncrowded in the patent landscape, with high commercial relevance for community ambulatory devices.
sEMG-Based vs. IMU-Based HIR: Key Dimensions
Click any row to explore further.
| Dimension | sEMG-Based HIR | IMU / Kinematic HIR |
|---|---|---|
| Signal Type | Muscle electrical activation (pre-movement) | Body kinematics, acceleration, joint angles |
| Peak Accuracy (dataset) | 92% with EMG + encoder fusion (2019) | 97.35% across 8 classes with deep learning (2022) |
| Latency Benefit | Captures intent before mechanical movement; adding EMG reduces latency by 27.1 ms vs. encoder-only | 0.506 s end-to-end latency reported; sub-10 ms inference on edge device |
| Inter-Subject Variability | High — requires individual calibration, limiting plug-and-play deployment | Lower — more robust across subjects and environments |
| Primary Applications | Upper-limb rehabilitation, lower-limb gait control, clinical settings | Industrial ergonomics, terrain classification, soft exoskeletons |
| Representative Patent | USTC sEMG intent assimilation control (CN, 2022, active) | IMU deep learning activity recognition (literature, 2022) |
| Key Limitation | Electrode placement sensitivity; signal noise and skin impedance variation | Does not capture pre-movement neural intent; relies on observable motion |
| Fusion Benefit | Adding EMG to encoder raises accuracy from 82.1% to 92% (2019) | BiLSTM fusion with force + displacement reaches 99.61% (2022) |
Frequently Asked Questions: Exoskeleton Human Intention Recognition
According to this dataset, the four core signal domains are: (1) biosignal-based sensing using sEMG and EEG; (2) inertial and kinematic sensing using IMUs, encoders, and force sensors; (3) novel contactless sensing using Hall effect sensors, magnetic fields, and camera-based vision; and (4) multi-modal fusion combining two or more of these modalities through machine learning.
In retrieved records, a 2022 study using BiLSTM fusion of force, displacement, and wheel speed sensor data achieved 99.61% recognition accuracy for a lower-limb rehabilitation robot. A separate 2022 study using IMU and encoder sensors with deep learning achieved 97.35% accuracy across 8 activity classes with 0.506-second end-to-end latency on an edge device.
High inter-subject variability is identified as the key limitation, requiring individual calibration and limiting plug-and-play deployment. This is noted in a 2019 study on detection of movement onset using EMG signals for upper-limb exoskeletons.
In this dataset, nine named assignees are identifiable: APTIMA Inc. (US), Free Bionics Taiwan Inc. (TW/US), Roam Robotics Inc. (US), Boston Dynamics Inc. (US), Hyundai Motor Company (KR), Massachusetts Institute of Technology (US), University of Science and Technology of China Institute of Advanced Technology (CN), University of Science and Technology of China (CN), and Mahendra Engineering College (IN).
The MIT WO 2026 patent on real-time anticipation and synchronization in close-proximity human-robot collaboration represents a paradigm shift from classifying current wearer intent to predicting future intent before movement occurs — enabling anticipatory robot response rather than corrective response. This is described in the dataset as a potential shift that could render reactive HIR architectures obsolete for high-performance applications.
The University of Science and Technology of China’s 2022 CN patent introduces an intent assimilation control method for upper-limb exoskeletons using sEMG signals. It uses a virtual target-based approach and introduces a continuous interaction spectrum spanning cooperative-to-competitive human-robot interaction modes, enabling nuanced physical guidance rather than binary assist or no-assist modes.
Data and insights on this page are based on a limited patent and literature dataset and are for reference only. Figures may not represent the complete technology landscape.