Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / How Robot Bots Are Learning to See With Touch
Transcript
- Lucas: Look, we have spent the last few weeks dissecting why robot arms still struggle with things like wet fish or loose cables, but there is a deeper layer here that we haven't fully unpacked yet. It isn't just about seeing the object clearly enough; it is about feeling what you are holding when your eyes lie to you. I want to talk about the sudden shift toward tactile sensing in industrial settings, because right now, as we sit here in mid-September twenty twenty six, the industry is moving hard from pure vision-based guidance to hybrid systems that actually use force feedback. Luna: That makes sense because vision can only tell you where something is, not how stiff or fragile it might be. You can see a glass jar, but you cannot know if it is empty or full until you pick it up. Lucas: Exactly. And that is the crux of the problem for anyone trying to automate delicate assembly lines. For years, the standard approach was to rely entirely on high-resolution cameras and deep learning models to predict grip strength based on visual texture. But in the real world, lighting changes, reflections happen, and sometimes an object is hidden behind another part in a bin. When vision fails, the robot either drops the item or crushes it because it has no idea how much force to apply. That is where tactile sensors start to pay off. Luna: So you are saying the data from the fingers is becoming just as important as the data from the eyes? That seems like a massive change in how these machines are programmed. Lucas: It is a fundamental architecture shift. We are seeing a move away from open-loop control, where the robot just executes a pre-programmed motion, toward closed-loop control where the system adjusts its grip in real time based on pressure readings. Think about a company like Tesla building their Optimus bots or even smaller players making cobots for pharmaceutical packaging. They need to handle vials that weigh less than a gram without cracking them. Vision alone cannot guarantee that level of precision under dynamic conditions. Luna: I remember reading about some early attempts with pressure-sensitive films in automotive assembly, but those were slow and messy to replace. How is the tech different now? Lucas: The difference is miniaturization and speed. The older technologies required bulky pads that couldn't fit into tight spaces and had refresh rates too slow for high-speed production. Today, we are looking at electronic skin materials and flexible sensor arrays that can be embedded directly into silicone fingertips. These new sensors can update hundreds of times per second, which is fast enough to prevent slippage before it even happens. It is basically giving the robot a sense of touch that rivals human sensitivity. Luna: But embedding all that hardware must add significant weight and complexity to the end of arm tooling. Does that hurt the speed of the overall operation? Lucas: It used to, but the latest generation of sensors is remarkably lightweight. We are talking about grams of added mass, not kilograms. The bigger challenge is actually the wiring and the signal processing. If you put ten thousand tiny sensors on a hand, you have a nightmare of cabling to manage inside a moving joint. Companies are solving this by integrating microcontrollers directly into the sensor mesh, sending digital signals over fewer wires instead of analog noise. It is elegant engineering, really. Luna: That sounds expensive though. Small manufacturers are already squeezing margins dry. Will they actually buy into this? Lucas: That is the million-dollar question. Initially, yes, the capital expenditure is higher because you are buying a more sophisticated gripper and a controller capable of processing haptic data. But the operational expenditure savings are compelling. If a robot can reduce waste by preventing crushed components, or if it can eliminate the need for expensive custom fixtures designed to hold parts in place, the return on investment kicks in much faster. We are seeing pilot programs in electronics assembly where defect rates dropped by forty percent after switching to tactile-enabled grippers. Luna: Forty percent is a huge number. That changes the math pretty quickly for any factory floor manager. Lucas: It does. And it opens up possibilities that were previously impossible. Take soft packaging, which we discussed a couple of episodes ago. Robots have always struggled with bags of chips or flexible pouches because they deform unpredictably. A vision system sees a blob. A tactile system feels the tension and can adjust its grip to slide along the surface without tearing it. This is why we are seeing more interest in logistics centers for e-commerce, where package shapes vary wildly every single day. Luna: I guess that explains why so many hardware startups are pivoting toward general-purpose manipulation instead of just picking and placing boxes. Lucas: Precisely. General-purpose manipulation requires dexterity, and dexterity requires feedback. You cannot manipulate a complex object if you cannot feel when you are starting to lose your grip. It is the same principle that allows humans to thread a needle while walking. We don't look at our fingers for every millimeter of movement; we feel the resistance. Robots are finally getting that capability, and it is changing the entire landscape of what they can do. Luna: It feels like we are moving from rigid automation to something more fluid. Does that make programming harder for the engineers? Lucas: It actually makes it easier in the long run, once you get past the initial setup. Previously, engineers had to teach the robot exactly where every part was located down to the millimeter. If a bin shifted by two centimeters, the whole process failed. With tactile sensing, the robot can explore the object, find its center of mass, and adjust its grip dynamically. It reduces the need for perfect part presentation and lowers the barrier to entry for deploying robots in unstructured environments. Luna: So the robot becomes more autonomous, less dependent on perfect factory conditions. That is a big deal for supply chains that are still recovering from recent disruptions. Lucas: Absolutely. Supply chains are less predictable now. Factories need to handle variety, not just volume. Tactile sensing allows for that flexibility. We are also seeing integration with AI models that learn from touch data. Instead of just reacting to pressure, the system builds a model of the object's physical properties over time. It learns that this specific plastic is slippery, that one is brittle, and adjusts its behavior accordingly. It is machine learning meeting mechanical engineering. Luna: I wonder if there are any safety benefits beyond just protecting the product. Maybe for working alongside humans? Lucas: That is a major driver for cobots, or collaborative robots. If a robot arm detects unexpected resistance, it can stop instantly. This prevents injury to workers who might accidentally step into the workspace. Traditional force limits are blunt instruments; they measure total torque. Tactile sensing is local. It can detect a collision on just one finger and stop that limb immediately while allowing others to continue. It creates a safer, more nuanced interaction zone. Luna: That kind of precision stopping could really change how companies design their factory layouts. Less fencing, more open space. Lucas: Right. It frees up floor space and reduces the need for costly safety retrofits. But let me pause here for a second to mention something about how we keep this show running. A handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that is literally what funds making this many of these detailed breakdowns possible. No ads, just listener support keeping the lights on. Luna: Yeah, it is great to have that direct connection with the audience. It lets us dig into the nitty-gritty without worrying about ad sponsors dictating the topics. Lucas: Agreed. Anyway, back to the tech. There is one more hurdle we need to address before we wrap up this angle. The cost of these sensors is coming down, but it is still a premium feature. Who is paying for it first? Luna: Likely the high-value industries. Pharmaceuticals, semiconductors, maybe luxury goods. Places where a single defective unit costs thousands. Lucas: Spot on. Semiconductors are a perfect example. Handling silicon wafers requires absolute care. One scratch ruins the batch. Vision helps, but touch confirms the grip. As the technology matures, we expect prices to drop, bringing it into consumer electronics assembly and eventually general warehousing. It is a classic adoption curve. Luna: Do you think this will eventually replace vision entirely, or will they stay paired? Lucas: They will stay paired. Vision is still superior for navigation and long-range targeting. You don't use touch to find a box across the room. But for the final act of grasping and manipulating, touch is indispensable. It is a complementary relationship. The future of robotics isn't just better eyes; it is better hands. And those hands are finally getting smart enough to feel. Luna: It is exciting to think about what other senses robots might develop next. Smell? Hearing? Lucas: Who knows? But for now, touch is the missing link that is unlocking so much potential. It turns robots from blind strongmen into precise artisans. We will see a lot more of this in the coming quarters as the hardware catches up to the software.