Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Industrial Robots Struggle With Compliant Tasks
Transcript
- Lucas: There is a fundamental disconnect in how we think about robot dexterity today. We measure success by how fast a machine can weld a car frame or pick up a steel ingot, but those are static, rigid objects. The real frontier isn't speed; it's compliance. That is the ability of a robotic arm to yield, to adjust its force in real time, rather than just moving to a precise coordinate. Luna: Right, because if you try to push a steel rod into a hole with pure position control, you will crush both the part and the robot. It is all about knowing how much pressure to apply. Lucas: Exactly. And this is where the industry is hitting a wall. For decades, industrial robots operated on position-based control. You tell the arm to go to point X, Y, Z, and it goes there. If there is resistance, the motor fights it until something breaks. But in any task involving flexibility—think fabric, cables, or even assembling fragile electronics—that approach fails completely. We need impedance control, which essentially makes the robot feel like it has joints that can breathe. Luna: So instead of fighting the environment, the robot adapts to it? Like a human hand wrapping around a glass without crushing it? Lucas: Precisely. Let me give you a concrete example that illustrates why this matters right now. Look at the textile and apparel sector. It accounts for roughly four percent of global manufacturing output, yet it remains one of the least automated industries. Why? Because fabric is not a box. It drapes, it wrinkles, it stretches. A traditional robot trying to fold a shirt using only cameras and position data will end up with a tangled mess every single time. The visual data tells it where the fabric is, but not how the fabric is behaving under tension. Luna: That sounds incredibly frustrating for engineers who have been trying to solve this for years. Is it just a software problem then? Lucas: It is a hardware-software co-design problem. You cannot solve compliance with code alone if your actuators are too stiff. To get true compliance, you need torque sensors embedded directly in the joints. These sensors measure the actual force being applied, not just the current sent to the motor. When you combine high-frequency torque sensing with an impedance control algorithm, the robot can detect a collision or a change in material stiffness within milliseconds and adjust its trajectory accordingly. Luna: I read that some newer collaborative robots are starting to include these joint torque sensors as standard. Is that changing the cost dynamic? Lucas: It is, but not as dramatically as you might hope. Adding high-precision torque sensors increases the bill of materials significantly. However, the real cost savings come from reduced downtime and fewer damaged parts. In a high-mix production line, where you switch products frequently, the ability to quickly tune the compliance parameters via software is worth more than the hardware cost itself. We are seeing early adopters in electronics assembly using this to insert connectors without snapping the pins. Luna: So it is less about replacing humans entirely and more about giving robots the sensitivity to work alongside delicate processes. Lucas: Yes, and this brings us to the broader implication for supply chains. As companies move toward reshoring or near-shoring manufacturing, labor costs are higher. Therefore, automation must be smarter, not just faster. If a robot can handle compliant tasks, it opens up entire categories of products that were previously deemed unautomatable. This includes medical device assembly, where you are handling tissue-like materials, or even food processing, where consistency in weight and shape is critical. Luna: Food processing is interesting because the margins are so thin. Would the ROI actually work out there? Lucas: It does, if you look at waste reduction. In poultry processing, for example, robotic deboning is a classic application. The meat is soft, irregular, and slippery. Older systems used rigid blades guided by vision, which resulted in high bone contamination rates. Newer systems use compliant grippers that can adapt to the shape of the carcass, improving yield by several percentage points. In a multi-billion dollar industry, that margin improvement pays for the technology very quickly. Luna: Yield improvements are always a strong sell for CFOs. But what about the programming side? Isn't tuning impedance control really complex? Lucas: Historically, yes. It required PhD-level control theory expertise. But the trend is toward abstraction. Manufacturers don't want to tune PID loops; they want to set a stiffness parameter. Think of it like the suspension in a car. You don't need to know the spring rate of every coil; you just choose a setting. Modern robot controllers are offering pre-set modes for different materials. Stiff for metal, soft for plastic, adaptive for unknowns. This democratization is key to adoption. Luna: That sounds much more accessible for the factory floor manager who isn't a controls engineer. Lucas: Exactly. And this ties into the concept of teach-by-example. Instead of coding every movement, you physically guide the robot through the task while wearing gloves that provide haptic feedback. The robot records not just the path, but the forces encountered. When it replays the task, it tries to match those force profiles. This closes the loop between human intuition and machine precision. Luna: Haptic feedback during teaching? That feels like it could be uncomfortable if the robot pushes back too hard. Lucas: Safety is built into the protocol. The system limits the maximum force transmitted to the operator. It is designed to feel like leading a horse, not dragging it. The goal is to make the learning curve flat. If a technician can teach a robot to assemble a circuit board in ten minutes using force guidance, rather than spending days writing code, that is a massive productivity gain. Luna: I am curious about the sensor density. Do we need many sensors, or can a few do the job? Lucas: A few well-placed sensors are often enough. Six-axis force-torque sensors at the wrist are the standard. They measure forces in three dimensions and torques around three axes. This gives you six degrees of freedom in sensing. For most assembly tasks, that is sufficient to detect misalignment and correct it. You do not necessarily need tactile skin over the whole arm, though that is an emerging field for more complex interactions. Luna: So the wrist sensor is the primary tool for compliance right now? Lucas: For general purpose industrial robots, yes. But we are seeing specialized end of arm tooling that integrates micro-sensors. For instance, a gripper with individual fingers that can sense slip or pressure independently. This allows for fine manipulation, like picking up a single grape without bruising it. It is still niche, but the technology is maturing rapidly. Luna: Picking a grape seems like a fun challenge, but I imagine the industrial applications are more serious. Lucas: Serious and lucrative. Consider the automotive industry's shift to electric vehicles. Battery pack assembly involves stacking thin, flexible layers of electrodes and separators. Any misalignment or excessive pressure can cause short circuits or fires. Compliance control is essential here to ensure uniform pressure across the stack. One major battery manufacturer reported a twenty percent reduction in defect rates after switching to compliant stacking robots. Luna: Defect reduction in battery manufacturing is huge given the safety implications. That is a compelling use case. Lucas: It is. And it highlights a pattern: compliance is becoming critical as products become more complex and fragile. We are moving away from heavy industry toward precision light industry. The robots need to be gentle. This also applies to pharmaceutical packaging, where blister packs require precise pressure to seal without damaging the pills inside. Luna: It seems like the definition of a 'robot' is expanding from a mechanical arm to a sensitive manipulator. Lucas: That is a great way to put it. And this shift is driving investment in new types of actuators. Traditional motors are powerful but stiff. Newer designs, like series elastic actuators, use springs in series with the motor to decouple the inertia. This makes the system inherently safer and more compliant. It mimics the muscle-tendon structure of biology, allowing for energy storage and shock absorption. Luna: Biologically inspired design again. It feels like nature has already solved these problems. Lucas: Nature has, but engineering constraints are different. We need repeatability and speed that biology doesn't always prioritize. However, the hybrid approach is promising. By combining rigid structures for power with compliant elements for interaction, we get the best of both worlds. This is the direction leading manufacturers like Universal Robots and KUKA are investing in heavily. Luna: Are these technologies available for small businesses yet, or is it still enterprise-grade? Lucas: The barrier to entry is dropping. Collaborative robots, or cobots, were designed for smaller footprints and easier setup. Now, they are incorporating compliance features that were once reserved for large articulated arms. A small contract manufacturer can rent a cobot with force feedback for a few thousand dollars a month. This lowers the risk of experimenting with automation for new product lines. Luna: Renting rather than buying makes sense for testing new processes before committing capital. Lucas: Absolutely. It changes the financial model from CapEx to OpEx. This agility is crucial in today's market where product lifecycles are shorter. You can reprogram a compliant robot for a new product in hours, whereas a dedicated fixture might take weeks to design and build. This responsiveness is a competitive advantage. Luna: I love that. Speed to market is everything right now. Lucas: And it all comes back to that core capability: the ability to interact with the physical world softly. We spent fifty years building robots that ignored friction and gravity. Now we are teaching them to respect them. Luna: That is a profound shift in perspective. From domination to collaboration with the materials themselves. Lucas: Precisely. And as these systems become more sophisticated, we will see robots taking on tasks that require not just precision, but judgment. Knowing when to push and when to yield. It is the difference between a machine and a tool. Luna: I wonder if this will lead to more jobs in robot maintenance and tuning rather than just operation. Lucas: Almost certainly. The demand for technicians who understand both mechanics and control theory is going to skyrocket. It is a new skill set for the trades. Not just fixing broken parts, but calibrating sensitivity and optimizing performance. Luna: So the future of work in manufacturing might be more technical than ever. Lucas: Yes, but hopefully less physically demanding. Reducing strain injuries is a major benefit of compliant automation. It creates a healthier workplace, which is good for retention and morale. Luna: Health and efficiency together. That is a strong combination for any business leader. Lucas: It really is. As we look at the next five years, compliance will move from a luxury feature to a baseline requirement for any advanced manufacturing cell. Luna: Makes sense. You cannot automate the modern world with rigid tools. Lucas: Agreed. The future is flexible, responsive, and surprisingly gentle. It is a quieter revolution, but one that will redefine what factories can produce. Luna: I am excited to see how this evolves, especially with the advancements in sensor technology we discussed. Lucas: Me too. It is a fascinating intersection of physics, software, and human-centric design. Luna: Before we wrap up, I want to mention that we deliberately do not run ads on these episodes. Lucas: If you want to support that choice and help keep the content independent, the link is buy me a coffee dot com slash fexingo. Luna: Your support directly funds the research and production, so thank you for listening and contributing. Lucas: Back to the topic, one final thought on where this is heading. We are starting to see AI models integrated directly into the compliance layer. Luna: AI predicting how a material will behave before the robot even touches it? Lucas: Exactly. Reinforcement learning allows the robot to simulate millions of interactions virtually, learning the optimal force profile for a specific material batch. It turns trial and error into predictive intelligence. Imagine a robot that knows exactly how much pressure to apply to a new type of composite because it has 'seen' it in simulation thousands of times. Luna: That removes a lot of the guesswork for engineers setting up new lines. Lucas: It does. And it accelerates deployment. What used to take weeks of tuning could be done in minutes. This is the next step in the evolution of industrial automation. Luna: From rigid machines to intelligent partners. It is a big leap, but it feels inevitable. Lucas: Inevitable. The question is no longer if robots will become compliant, but how quickly we can scale that capability across every industry. Luna: A worthy goal for the coming decade. Thanks for diving into this with such depth. Lucas: Thanks, Luna. It is a complex topic, but essential for understanding the next wave of manufacturing innovation.