Posted in

What are the torque – control strategies for robot joint actuators?

If you’re deep in the robotics space, you know robot joints make or break a bot—whether it’s a collaborative robot assembling electronics, a mobile robot navigating a warehouse, or an exoskeleton helping someone move. At the core of every good joint is its actuator, and the biggest thing we deal with as a supplier is getting torque right. Mess up torque control, and you end up with a bot that’s too weak to lift a part, too jittery to work safely around humans, or wastes half its energy heating up unnecessarily. Over the years, I’ve walked through dozens of customer projects where torque control was the make-or-break detail, so I want to break down the key strategies we rely on here at our shop, and why they matter for your next build. Robot Joint Actuators

First, let’s get on the same page: what’s torque control, exactly? Unlike position control, which tells an actuator “move to this angle,” or velocity control, which says “spin at this speed,” torque control is all about “apply this exact amount of twisting force right now.” For a cobot lifting a delicate circuit board, that means not squeezing too hard and damaging the part. For an industrial robot arm moving heavy packages, that means not so weak it drops the load, but not so strong it wears out the joint in a year. It’s a balance that’s way trickier than it sounds, and that’s why different strategies exist for different jobs.

One of the most common basic strategies is Proportional-Integral-Derivative (PID) torque control. You’ve probably heard of PID for position or velocity, but it works here too. Here’s how it plays out for our actuators: we set a target torque (say, 20 Nm for a pick-and-place task), and the PID controller compares that target to what the actuator is actually outputting, then adjusts the motor’s voltage in real time to close the gap. The proportional term gives immediate correction if the torque is off, the integral term fixes any long-term drift (like if friction makes the actuator apply 1 Nm less than it should over time), and the derivative term dampens jitter so the joint doesn’t shake. That said, PID isn’t one-size-fits-all. We’ve seen customers use it successfully for simple fixed tasks, but it can struggle with sudden changes—like a cobot grabbing a part that’s heavier than it expected, or a mobile robot climbing a curb where load changes fast. That’s why most of our custom actuator builds tweak this, not just use off-the-shelf PID.

Next up, and super important for safety-focused robots (like cobots or exoskeletons), is current-based torque control. Wait, why current? Because for most permanent magnet DC motors (which are the workhorse for small to medium robot joints), the amount of twisting force (torque) is directly proportional to the current going through the motor windings. That’s a physical law we can count on, right? So instead of measuring torque directly (which requires expensive torque sensors), we just measure the current, set a target current, and adjust the motor to hit it. The big win here is cost: torque sensors are pricey, bulky, and add weight to the joint, which for drones or wearable exoskeletons is a non-starter. We use current-based control for 70% of our off-the-shelf actuators, especially for cobots that need safe, consistent torque without the extra bulk. The catch? It’s indirect. If there’s a flaw in the motor or the gearing, the current might not perfectly translate to torque—so we always add a small calibration step when we ship actuators, matching current readings to actual torque output for the specific gearing the customer uses.

For more precision and dynamic tasks, like a robot arm assembling tiny components or a surgical robot, we lean into direct torque control (DTC). This one’s a bit different because it actually measures torque directly with a small, high-resolution sensor mounted right on the joint output, not the motor. The controller then uses that real torque reading to adjust the motor, instead of inferring it from current. The response time is way faster—microseconds, instead of the milliseconds you get with current-based or PID. That’s a game-changer for applications where even a tiny over-torque can ruin a part or hurt someone. Last year, a medical device customer came to us because their surgical robot kept slipping when gripping small tissue samples; switching to DTC fixed the issue, and they cut their part defect rate by 12%. The tradeoff is that DTC is more expensive and a bit larger, so we only recommend it when the job needs that level of precision. It’s not worth it for a warehouse robot moving pallets, but for delicate, high-stakes tasks, it’s the go-to.

Then there’s feedforward torque control, which is all about pre-empting torque needs instead of reacting to them. Think of it like driving a car: if you’re approaching a hill, you don’t just press the gas when you start slowing down—you press it a little early, so you don’t lose speed. For robot joints, feedforward uses a model of the joint’s load, friction, and inertia to calculate what torque the actuator will need before it even moves, then adds that to the PID or current-based control. We use this a lot for moving parts with predictable loads, like linear actuators or industrial robot arms doing repeated pick-and-place cycles. For example, if a robot arm always lifts a 5 kg box to the same height every 10 seconds, the feedforward controller will “remember” the torque needed for that movement and dial it in instantly, instead of waiting for the feedback controller to correct it. This reduces wear on the actuator too, because it’s not constantly adjusting to over- or under-torque, which extends the lifespan of the gears and motor by about 20% in our customer testing.

Now, what about cases where loads change constantly—like a mobile robot navigating uneven terrain, or a humanoid robot interacting with its environment? That’s where adaptive torque control shines. Unlike the other strategies that use a fixed model, adaptive control learns and adjusts to changes in real time. If a humanoid robot steps on a rock, the adaptive controller detects the unexpected torque spike, updates its model of the load, and adjusts the actuator’s output within milliseconds. We’ve worked with a humanoid robot maker that had a lot of trouble with their prototype tripping and applying too much torque to their knee joints, which damaged parts every few weeks. Swapping in our custom adaptive torque control actuators fixed that—now the robot adjusts its torque mid-step without any programming changes. The downside here is complexity: adaptive algorithms require more processing power, so we have to pair these actuators with slightly more capable controllers, which adds a little cost. But for dynamic, unstructured environments, it’s hard to beat.

Wait, let’s talk about a common mistake we see new customers make: relying on a single torque control strategy for everything. We had a startup a couple of years ago that wanted to build a cobot for both packaging heavy pallets and assembling tiny electronics. They ordered our standard current-based actuators, thinking it would work for both. For the pallets, the torque was too inconsistent (current measurement didn’t account for gearing friction that changed with load), and for the electronics, it was too slow to adjust when gripping tiny parts. We ended up modifying their order: we kept the current-based control for the pallet arm, and added a DTC module for the end effector that does the electronics assembly. That split strategy worked way better than trying to force one strategy to fit both jobs. It’s a reminder that the right torque control isn’t just about picking a strategy—it’s about matching it to your specific use case.

As a joint actuator supplier, we also spend a lot of time on calibration when we ship units, because even the best strategy won’t work if the actuator isn’t calibrated correctly. For current-based systems, we test each actuator with different loads, mapping current readings to actual torque, so when the customer sets a target, it’s accurate within 2%. For DTC systems, we align the torque sensor right on the output shaft to avoid measurement errors, which is a detail we don’t cut corners on. We also let customers customize control parameters for their specific application—if someone’s building a robot that works in extreme cold, we can adjust the integral term to account for extra friction from cold lubricants, for example.

Now, let’s wrap this up with what matters most to you: why does all this torque control stuff matter for your robot? The right strategy means your bot is safer (no over-torque that can pinch humans), more reliable (less wear and tear on parts, fewer breakdowns), and more efficient (less energy wasted on unnecessary torque, which translates to longer battery life for mobile robots). If you’re planning a new build, don’t just go with whatever actuator comes with your controller—talk to someone who knows torque control for joints.

If you’re looking for high-quality robot joint actuators tailored to your torque control needs, reach out to our team to discuss your project. We’ve worked with cobot makers, industrial robot teams, medical device developers, and more, so we can help you pick the right strategy, calibrate it for your use case, and get actuators that perform when you need them.

Humanoid Robot Skeleton References

  1. Craig, J. J. (2005). Introduction to Robotics: Mechanics and Control. Pearson Education.
  2. De Wit, C. C., Siciliano, B., & Bastin, G. (1996). Theory of Robot Control. Springer.
  3. El-Hawwary, M. I., & Maguire, C. (2018). Direct Torque Control of Permanent Magnet Synchronous Motors: A Review. IEEE Transactions on Power Electronics.
  4. Hori, Y., Sawada, H., & Chun, Y. (1991). Sensorless Torque Control of Induction Motor Based on Magnetizing Current Vector Control. IEEE Transactions on Industrial Electronics.
  5. Lewis, F. L., Abdallah, C. T., & Dawson, D. M. (1993). Control of Robot Manipulators. Macmillan Publishing Company.
  6. Visioli, A. (2006). Practical PID Control. Springer.
  7. Zhang, X., & Wang, Y. (2020). Adaptive Torque Control for Robots Interacting With Uncertain Environments. International Journal of Robotics Research.

Jiangsu Zhengfang Dynamics Technology Co., Ltd.
As one of the most professional robot joint actuators manufacturers and suppliers in China, we’re featured by quality products and good price. Please rest assured to buy customized robot joint actuators made in China here from our factory. Contact us for pricelist.
Address: Building 3, No. 69 Feitian Avenue, Jiangning District, Nanjing City, Jiangsu Province
E-mail: hanks.liu@zfdynamics.com
WebSite: https://www.zhengfangdongli.com/