Ryan holds a BS and MS in Mechanical Engineering from the Pennsylvania State University. He has 10 years of industry experience in electro-mechanical sensor design and manufacturing.
Ryan joined HBM (now HBK) 9 years ago as a Product Development Engineer and has managed the OEM Sensors engineering team since 2018.
He has spent his entire career working directly with customers in industries such as automotive, robotics, medical, and agriculture to provide highly customized solutions for an array of unique challenges.
Following his presentation at Robotics Summit Boston, “Designing Production-Ready Force and Torque Sensors for the Next Generation of Robots”, Ryan Young, Global Head of Engineering OEM Sensors at HBK, engaged with a wide range of robotics experts, OEMs, and industry leaders to discuss the evolving challenges of physical interaction. In this context, we sat down with him to capture key insights shaped both by his expertise and by the discussions held at the event.
He shares why robotics companies are rethinking force and torque sensing, how customised 6‑DoF solutions are emerging as a new source of competitive advantage, and why sensing quality could become a foundational element of Physical AI.
For decades, competitive advantage came from making machines move faster, more precisely, and more reliably. That was sufficient when robots operated within structured, repeatable environments.
Today, those environments have fundamentally changed. A humanoid robot climbing stairs must continuously interpret ground reaction forces to stay upright. A collaborative robot for assembling electronics must detect subtle resistance before a component fails. A warehouse manipulator must distinguish between a secure grasp and a slipping object – not after the fact, but at the moment it occurs.
Knowing where a robot is located is no longer sufficient. It must understand how it is physically interacting with its surroundings.
This is the shift from motion execution to physical interaction. Sensors in these systems are not passive recorders; they are active participants in real-time decision-making. That is what makes force and torque sensing a strategic technology rather than an instrumentation detail.
Off-the-shelf sensors are built on a reasonable assumption: that most applications share enough requirements to justify a common solution. In robotics, that assumption rarely holds true for long.
A humanoid balancing on uneven terrain, a surgical robot operating inside the human body, and a warehouse manipulator handling thousands of different products all face fundamentally different mechanical, environmental, and control challenges. As robots become more advanced, those differences become increasingly significant.
Teams that begin with catalogue sensors rarely encounter a single limitation. Instead, they encounter a series of trade-offs. A sensor with the right force range may add too much mass. One with the right mass may introduce unacceptable stiffness. Another, while offering acceptable stiffness, may lack the multi-axis accuracy required by the control system.
Each compromise creates another workaround – a mechanical adaptation, a software filter, a recalibration routine, or a change elsewhere in the system.
The complexity does not disappear. It accumulates, often invisibly, in other parts of the project. By late-stage development, a significant portion of the engineering effort may be spent compensating for a sensing solution that was never designed for the application.
As a result, leading robotics companies are increasingly asking a different question. Instead of asking, “Which sensor should we buy?”, they ask, “What sensing architecture best supports the robot we are building?”
This shift changes where sensing sits in the design process – from a procurement decision made late in development to a foundational architectural decision made at the very beginning.
A 6-DoF force and torque sensor is unique because it sits at the intersection of mechanics, electronics, software, and control.
Its dimensions affect packaging. Its mass influences robot dynamics. Its stiffness impacts force control performance. Its latency affects control loop stability. Its communication interface influences system integration.
Its placement within the robot is equally important. A wrist-mounted sensor on a collaborative robot serves a different purpose than a joint-integrated sensor in a humanoid ankle, or a force sensor embedded within a robotic gripper.
In other words, the sensor does not simply measure interaction. It directly influences robot behaviour.
This is why advanced robotics companies increasingly view sensing as part of the overall system architecture rather than as a stand-alone component.
The most successful projects optimise the robots and sensing solutions together.
Strain gauges remain the reference standard for force and torque sensing in applications that demand accuracy, stability, and long-term repeatability. Unlike indirect methods that infer forces through models or secondary effects, strain gauges directly measure the mechanical deformation of a structure under load. What you get is a measurement of what is physically happening, not an estimate.
Most engineers understand that in principle. What they underestimate is how much of the real challenge lies downstream of the gauge itself: mechanical structure design, gauge placement and orientation, signal conditioning, multi-axis decoupling, calibration, thermal compensation. Each layer introduces potential error. A signal with drift, hysteresis, or axis crosstalk does not just reduce measurement accuracy in a test environment, it actively degrades the performance of every system that depends on it.
A force control loop built on noisy data oscillates. An AI model trained on biased measurements learns incorrect physical behaviour and that bias is baked in, not correctable in software after the fact. A safety system relying on inaccurate contact detection responds to phantom events or misses real ones.
High-quality physical data does not begin with the AI layer. It begins with the measurement chain.
At HBK, measurement science is the foundational discipline. The engineering challenge is not integrating a strain gauge into a housing, it is designing the entire chain, from mechanical structure to calibration algorithm, to deliver data that accurately reflects what is happening at the point of contact.
Engineers often focus first on measurement range, but in practice, many other parameters can be equally important.
Sensor height can influence workspace and payload capacity
Sensor mass affects dynamic performance and energy consumption
Stiffness influences interaction quality and control stability
Latency determines how quickly a robot can react to unexpected events
Crosstalk directly impacts multi-axis force accuracy
Communication protocols affect integration effort and software architecture
For example, in a high-speed pick-and-place application, reducing sensor mass by a few hundred grams can significantly improve cycle times and reduce actuator loads. In a humanoid robot, sensor stiffness and latency can directly affect balance control and walking stability. In medical robotics, accuracy and repeatability may take priority over payload capacity.
What makes robotics challenging is that these parameters are highly interconnected. Optimising one often affects another. The best solution is rarely the one with the highest specification in a single category, it is the one that delivers the best balance across the entire system.
It is a reasonable assumption, and it is often wrong.
When a standard sensor cannot fully meet system requirements, the complexity does not disappear: it shifts. Engineers redesign brackets to accommodate the form factor. They develop software filters to manage noise characteristics. They accept performance compromises that require additional validation rounds later in the process. These workarounds are carried out throughout the entire development programme.
By the time a team reaches late-stage integration, the cost of those accumulated compromises often exceeds what a purpose-built solution would have required at the start – when changes were cheap.
A custom-designed sensing solution concentrates engineering effort at the beginning of the project. The mechanical structure, electronics, calibration methodology, and communication architecture are all developed around the robot’s actual requirements from day one. The result is a cleaner system with fewer downstream surprises and a more straightforward path to validation and scaled manufacturing.
Custom sensing is not product customisation. It is system optimisation and it is most valuable when started early.
One programme involved an advanced manipulation platform targeting high-volume industrial deployment. The requirements were clear: a sensor footprint under 15 mm in height; electronics fully integrated into the sensor body; six-axis accuracy with less than 1% crosstalk across all axes; latency under 1 ms; mechanical robustness to withstand the shock and vibration profile of the application; and a unit cost compatible with production volumes in the tens of thousands.
Every catalogue product evaluated could meet two or three of those requirements. None could meet all of them simultaneously – particularly the combination of compact integrated electronics, sub-millisecond latency, and the target cost at scale.
By co-designing the mechanical structure, strain gauge layout, conditioning electronics, calibration approach, and manufacturing process as a single system, it was possible to satisfy all six constraints within the programme timeline. The sensor did not exist as a stand-alone product; it existed as a solution to a specific robot’s specific requirements.
The outcome was a platform capable of operating in applications – and at cycle times – that the team had previously considered out of reach. In a market where the performance envelope of a robot directly determines which contracts it can win, this translates into a durable commercial advantage.
Humanoid robotics concentrates almost every sensing challenge into a single platform, simultaneously.
The form factor demands compact, low-mass sensors that can be integrated into ankles, wrists, and fingers without compromising range of motion or adding weight that degrades locomotion efficiency. The mechanical environments at each location are entirely different: ankle sensors must absorb impact loads during dynamic walking while providing accurate balance data; wrist sensors must support dexterous force-controlled manipulation; finger-integrated sensors must function within millimetre-scale constraints while delivering meaningful force resolution.
The control requirements are equally demanding. Bipedal balance requires continuous low-latency ground reaction force data. Object manipulation requires precise multi-axis force regulation. Safe human interaction requires reliable detection of unexpected contact anywhere on the body.
This is one reason demand for customised 6-DoF force and torque sensing is accelerating. Humanoid systems need more than contact detection; they need to understand forces and moments across all six degrees of freedom to support balance, manipulation, and safe interaction with the physical world.
And all of this must be manufacturable at a cost that makes commercial deployment viable, not just technically achievable in a lab.
What makes humanoid development significant for the broader industry is that it forces every trade-off into the open at once. The sensing solutions being developed for humanoid platforms are already influencing logistics automation, collaborative robotics, and mobile manipulation applications that share many of the same constraints.
The robotics AI community is investing heavily in model architecture, training methodologies, and simulation-to-real transfer. These are genuinely important challenges. However, they share a dependency that often receives less attention: the quality of the physical data the models learn from and act on.
An AI model learning manipulation from force and torque data learns about the physical world as its sensors represent it. Drift, noise, crosstalk, and latency in the sensing layer do not simply reduce measurement accuracy; they introduce systematic distortions into the model’s representation of physical reality. That distortion is not easily corrected later. It is learned.
Conversely, high-fidelity physical data- – accurate, low-latency, well-calibrated 6-DoF measurements –provides a learning system with a richer and more reliable foundation. Skills learned from clean physical data tend to generalise better to new objects, surfaces, and environments than those learned from noisy or biased inputs.
Force and torque sensors are becoming a foundational data layer for Physical AI. The quality of that layer shapes what a model can learn, not just how accurately it can act at a given moment.
The practical implication is that sensing quality decisions made during robot design have consequences that extend throughout the entire lifecycle of the AI system running on that robot. It is not a hardware detail to be resolved downstream of model development. It is a design variable that helps determine the upper limits of what the system can achieve.
Download the “Custom Sensor Assembly for OEM” guide to learn how to design, prototype, and scale high-quality sensors. Gain expert insights, speed up time-to-market, and optimize production to create innovative solutions that meet customer needs and boost your competitive edge.
For many years, competitive advantage in robotics was driven primarily by mechanical design.
More recently, software and artificial intelligence became major differentiators.
The next competitive frontier will be physical intelligence.
The most successful robots will not simply move more accurately. They will understand contact, adapt force in real time, react safely to uncertainty, and operate effectively in environments that cannot be fully predicted.
Achieving this requires more than advanced AI.
It requires sensing systems specifically designed around the physical realities of the application.
The companies that recognise sensing architecture as a strategic design decision today will be the ones defining the future of robotics tomorrow.
Ultimately, the winners will be those capable of combining mechanical excellence, intelligent software, and high-quality physical data into a single, optimised robotic platform.
That journey begins with understanding that sensing is no longer just a component.
It is becoming a strategic enabler of robot intelligence itself.
Standard sensors often reach their limits when faced with today’s engineering challenges. With HBK, you can design and produce sensors that combine precision, performance, and seamless integration—engineered to fit perfectly into your application.
This will bring together HBM, Brüel & Kjær, nCode, ReliaSoft, MicroStrain and Discom brands, helping you innovate faster for a cleaner, healthier, and more productive world.
