Cobot Force Torque Control Landscape 2026 — PatSnap Eureka
Collaborative Robot Force Torque Control: Patent & Innovation Landscape 2026
Sensor architectures, control algorithms, and safety frameworks are converging to enable cobots to operate safely alongside humans. This report maps 14 patent families and 30+ literature sources spanning ISO/TS 15066-compliant force control from foundational 1990 filings through active 2026 applications.
Three Subsystems Define Cobot Force-Torque Control
Collaborative robot force-torque control is defined by the intersection of three technical subsystems: sensing mechanisms that measure joint torques, end-effector forces, or estimate them virtually; control architectures that translate those measurements into safe, compliant motion commands; and safety and response logic that detects collisions, classifies intent, and triggers appropriate robot behavior.
The field is grounded in the principle that robotic joints must either physically measure or computationally estimate forces and torques to enable compliance. Early patents such as the Westinghouse Electric multiaxis robot controller (US, 1992) established the foundational concept of controlling workpoint torque as an end variable rather than an intermediate variable — a distinction that remains central to modern cobot design. Contemporary filings extend this with neural-network-based dynamic modeling, impedance control, and intent-recognition frameworks driven by force/torque data streams.
Among the retrieved results, 14 distinct patent records and over 30 literature sources span these subsystems across jurisdictions including US, EP, JP, CN, FR, and WO. Sub-domains identified include joint torque sensing and virtual sensing, collision detection and mode-switching control, impedance and admittance control, force-based task learning and imitation, multi-robot torque-coordinated cooperation, and safety threshold management under ISO/TS 15066. For broader context on human-robot collaboration standards, the IEC and IEEE publish complementary safety and communication frameworks. PatSnap’s IP analytics platform enables landscape analysis across all these domains.
From Foundational Torque Control to AI-Integrated Cobots
Patent activity spans three distinct eras — each building on the prior and accelerating toward AI-integrated force sensing architectures.
Workpoint Torque as a Controlled Variable
Staubli International AG (EP, 1990) and Westinghouse Electric (US, 1992) established the concept of controlling workpoint torque as an end variable rather than an intermediate variable — a distinction central to modern cobot design. Kabushiki Kaisha Toshiba’s Robot Control Apparatus (US/EP, 2013) introduced external torque estimation through the difference between estimated drive torque and commanded torque, a method still widely applied. Nachi-Fujikoshi’s articulated robot torque control work appeared in Japan in 2007.
Foundational concept still in useApplication-Specific Proliferation
General Motors filed force-torque imitation learning patents in CN (2015, 2017). X Development LLC patented programming and execution of force-based tasks with torque-controlled robot arms (US, 2015). CEA filed an H∞-optimized force amplification control patent (FR, 2017). Hanwha Robotics produced a collision detection mode-switching patent family in US and EP (2020). BAE Systems filed a torque-threshold management system across EP (2019) and US (2022). Shanghai Jiao Tong University filed a torque-sensor-based integrated drive-control architecture (CN, 2018).
Application-specific proliferationNeural Networks, Precision Sensing, Multi-Robot Coordination
Schaeffler Technologies AG filed a method for precisely determining output torque (US, 2026 — the most recent patent in this dataset). Intel Corporation filed a neural-network cobot model generation system (EP, 2025). Nachi-Fujikoshi filed a safety-tiered escape mode system (JP, 2025; US, 2024). Jaka Robot filed a torque inner-loop multi-robot coordination method (CN, 2025). UBTECH Robotics filed a Riemannian Motion Policy-based force control method (US, 2025). Active and pending filings from 2022–2026 constitute approximately 60% of total patent records in this dataset.
~60% of dataset from 2022–2026Highly Fragmented: 14 Distinct Assignees Across 14 Records
Innovation is not concentrated in a few players — 14 distinct assignees appear across 14 patent records, suggesting a highly fragmented competitive field. Chinese academic institutions and robotics startups are among the most active recent filers, while established European industrial companies such as Schaeffler and CEA continue to contribute precision control innovations. This fragmentation signals an open competitive landscape with significant whitespace for new entrants, particularly in multi-robot force coordination.
Highly fragmented fieldFour Control Approach Clusters in the Patent Dataset
Patent and literature evidence clusters around four distinct technical approaches to cobot force-torque control, from hardware sensing to AI-driven intent recognition.
Filing Activity by Era
Active and pending filings from 2022–2026 constitute approximately 60% of total patent records, signalling accelerating innovation.
Cluster Representation in Dataset
Physical joint torque sensing is the dominant approach; AI-driven intent recognition represents the leading edge of active filings.
Control Architecture Progression: From Hardware to AI
The four clusters follow a logical progression from physical sensing through virtual estimation to fully AI-driven adaptive control.
Where Cobot Force-Torque Control Is Being Deployed
| Domain | Key Patent / Source | Assignee | Year | Force Control Role |
|---|---|---|---|---|
| Industrial Assembly | Programming and Execution of Force-Based Tasks | X Development LLC | 2015 | Joint torque commands track position and exert prescribed end-effector forces simultaneously |
| Industrial Assembly | RAM Insertion via Reinforcement Learning (Literature) | — | 2021 | Torque-controlled robot performs contact-rich assembly using proactive visual residual RL |
| Medical / Surgical | Automatic Selection of Collaborative Robot Control Parameters | Koninklijke Philips N.V. | 2023 | Distinguishes intentional user guidance forces from accidental contact in surgical procedures |
| Medical / Surgical | Computed Torque Control with RBF Neural Networks (Literature) | — | 2020 | Sub-millimeter force-regulated positioning for percutaneous puncture surgery |
US Leads in Volume; China Is the Fastest-Growing Origin
Among the 14 patent records with assignee data in this dataset, the United States is the largest single jurisdiction with 10 filings, including active grants and pending applications. Key US assignees include X Development LLC (Google), Hanwha Robotics Corporation, BAE Systems PLC, Koninklijke Philips N.V., Neuromeka, Nachi-Fujikoshi Corp., Schaeffler Technologies AG, Southeast University, UBTECH Robotics Corp., and Dexterity Inc.
China has 8 CN-jurisdiction records, representing the fastest-growing filing base in this dataset. Key CN assignees include Shanghai Jiao Tong University, Ningbo Institute of Materials Technology and Engineering (Chinese Academy of Sciences), Jaka Robot Co., Ltd., KUKA Robot Manufacturing (Shanghai), Tencent Technology (Shenzhen), and Beihang University Hangzhou Innovation Research Institute. IP strategists should treat CN as a primary jurisdiction — freedom-to-operate analysis must be China-first.
Japan has 3 JP-jurisdiction records, with Nachi-Fujikoshi and Yaskawa Electric as primary filers, reflecting Japan’s long-standing position in industrial robot hardware design. Europe includes BAE Systems (EP, 2019), Hanwha Robotics (EP, 2020), Staubli International (EP, 1990), CEA (FR, 2017), and Intel Corporation’s EP filing (2025) — a notable entry of a major semiconductor company. For global patent filing trends, WIPO and EPO publish complementary jurisdiction data. PatSnap’s materials and engineering solutions and customer case studies demonstrate how IP teams use landscape analysis to prioritize jurisdictions.
Five Signals Visible in the Most Recent Filings
The 2024–2026 filing cohort reveals directional bets being placed by established industrials, academic spinouts, and AI-native companies.
Precision Sensorless Output Torque Determination
Schaeffler Technologies’ 2026 US pending application directly addresses the gap between expensive dual-sensor systems and imprecise motor-current-based estimation, targeting high-performance cobot systems without hardware overhead.
Neural-Network Cobot Model Generation
Intel Corporation’s 2025 EP filing uses a trained neural network to map joint trajectories from a generic robot model to a specific cobot model, enabling generalized force-torque control transfer across cobot morphologies.
Tiered Safety and Escape-Mode Logic
Nachi-Fujikoshi’s 2024 US and 2025 JP filings implement multi-threshold collision response logic that distinguishes between worker entrapment scenarios requiring escape mode activation versus heavy payload scenarios where escape mode would be dangerous — a significant advance in ISO/TS 15066 implementation granularity.
Torque Inner-Loop Multi-Robot Coordination
Jaka Robot’s 2025 CN filing distributes torque-based compliance control across master and slave robots using preset direction matrices, eliminating the need for a high-performance central computing core and lowering the hardware barrier for multi-cobot deployment.
Five Decisions IP and R&D Teams Must Make Now
The competitive and technical signals in this dataset have direct implications for product development, IP strategy, and freedom-to-operate analysis.
Sensor Hardware vs. Software Estimation Is the Defining Choice
Multiple active filings — Schaeffler 2026, virtual force sensor literature 2023, Shanghai Jiao Tong University 2018 — converge on the challenge of achieving hardware-grade torque accuracy through software means. R&D teams entering this space must choose a position on this spectrum early, as it constrains downstream control architecture choices. PatSnap’s IP analytics tools can map the prior art landscape for each approach.
Architecture-constraining decisionISO/TS 15066 Compliance Is Driving Hardware-Software Co-Design
Nachi-Fujikoshi’s escape-mode logic and BAE Systems’ threshold iteration system both embed safety-standard requirements directly into the control law. IP strategists should monitor whether this safety-by-design approach becomes a mandatory qualification barrier in target markets. Relevant standards context is available from ISO and IEC.
Potential qualification barrierChina Is an Innovation Origin, Not Just a Manufacturing Destination
With filings from Chinese Academy of Sciences, Shanghai Jiao Tong University, Jaka Robot, Tencent, KUKA Shanghai, and Beihang University all appearing in this dataset, product developers and IP strategists cannot treat CN as a secondary jurisdiction. Freedom-to-operate analysis must be China-first. The WIPO PCT database and PatSnap’s API access support comprehensive CN prior art searches.
China-first FTO requiredAI-Integrated Force Control Expected to Be Table-Stakes by 2027–2028
Koninklijke Philips’ intent-recognition system, GM’s force-torque imitation learning, and Intel’s neural-network cobot model all embed machine learning directly into the force-torque control loop. Teams building next-generation cobots should expect AI-integrated force control to become a table-stakes feature by 2027–2028. Research context is available from IEEE robotics publications.
Table-stakes by 2027–2028Collaborative Robot Force Torque Control — key questions answered
Collaborative robot force-torque control encompasses the sensor architectures, control algorithms, and safety frameworks that enable robotic arms to sense, regulate, and respond to physical forces during human-robot interaction. It is defined by three technical subsystems: sensing mechanisms, control architectures, and safety and response logic.
The four main clusters are: (1) Physical joint torque sensing and threshold-based safety control, (2) Virtual sensing and sensorless estimation, (3) Impedance/admittance and force-based task execution, and (4) Force/torque-driven intent recognition and adaptive control parameter selection.
The United States is the largest single jurisdiction with 10 filings. China has 8 CN-jurisdiction records and represents the fastest-growing filing base. Japan has 3 records, with Nachi-Fujikoshi and Yaskawa as primary filers. Europe includes BAE Systems, Hanwha Robotics, Staubli, CEA, and Intel.
Active and pending filings from 2022–2026 constitute approximately 60% of total patent records in this dataset.
The defining tradeoff is between physical sensor hardware and software-based estimation. Multiple active filings converge on the challenge of achieving hardware-grade torque accuracy through software means. R&D teams must choose a position on this spectrum early, as it constrains downstream control architecture choices.
Five directional signals are visible: (1) Precision sensorless output torque determination, (2) Neural-network cobot model generation, (3) Tiered safety and escape-mode logic, (4) Torque inner-loop multi-robot force coordination without centralized computation, and (5) Riemannian Motion Policy-integrated force control.
PatSnap Eureka searches patents and research literature to answer instantly.