Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / How Robots Learn to See With Touch
Transcript
- Lucas: If today's tech conversation gave you something usable, I appreciate it. And if you want to keep these deep dives coming without ads cluttering up your feed, a couple of dollars a month is genuinely what keeps these going. You can support us at buy me a coffee dot com slash fexingo. Thanks for listening. Luna: Yeah, that really does help keep the lights on in the studio. Let's get back to the lab. Lucas: Right. So here is the thing about industrial robotics that most people miss. For the last thirty years, we have been building machines that see everything but feel nothing. We put cameras on robotic arms in warehouses and factories, and we told them to pick up a box, fold a shirt, or stack a pallet. The cameras could locate the object with millimeter precision, but they had no idea what the object felt like. Luna: So the robot knows where the item is, but not how fragile or slippery it actually is until it touches it? Lucas: Exactly. And that distinction has become a massive bottleneck as we move into mid twenty twenty six. Visual systems are fantastic for rigid, predictable geometry. Think of an automobile assembly line where a car door is always the same shape and weight. But now we are trying to automate warehouses that handle clothes, fresh produce, and irregular packaging. Those things change shape under pressure. A camera cannot tell you if you are crushing a tomato or just holding it firmly. Luna: It sounds like we are finally moving past the era of purely visual automation. Lucas: We are. And the companies leading that charge right now are shifting their R&D budget heavily toward tactile sensing. Look at what Figure AI is doing with their humanoid platforms. They are embedding thousands of microscopic force sensors directly into the fingertips of their robotic hands. It is not just about grip strength anymore. It is about real-time haptic feedback loops that allow the robot to adjust its pressure every few milliseconds. Luna: That is a significant leap from the simple on-off switches we saw in earlier generation grippers. Lucas: Huge leap. In the old days, a robot would clamp down with a fixed percentage of its maximum capacity. If it picked up a bag of chips, it might crush it because the algorithm didn't know the bag was empty and therefore weak. Now, with dense tactile arrays, the robot feels the resistance. If the surface yields too quickly, it reduces torque instantly. It mimics the way human skin tells our brain to stop squeezing. Luna: But isn't adding all those sensors going to drive up the hardware cost significantly? Lucas: It used to be prohibitive. That was the exact problem three years ago. But the component supply chain has matured rapidly. The cost of flexible electronic skins and piezoresistive sensor pads has dropped by nearly forty percent since early twenty twenty four. This makes it economically viable to outfit even entry-level warehouse bots with basic tactile awareness. Luna: So the economics are finally catching up to the engineering possibilities. Lucas: Precisely. And this is where the operational impact becomes visible. Take a major logistics provider like Amazon or FedEx. They are testing these tactile-enabled arms in their sortation centers. The goal is to reduce damage rates for non-rigid goods. When a robot can sense that a package is bulging or made of thin plastic, it adjusts its approach angle and grip force automatically. This reduces the need for manual inspection teams later in the chain. Luna: That seems like a direct correlation between better sensors and lower overhead costs. Lucas: It is. And it goes beyond just preventing damage. It enables new types of automation that were previously impossible. Consider the garment industry. Folding shirts is notoriously difficult for machines because fabric drapes and shifts unpredictably. Visual only systems struggle to identify the edges of a folded shirt once it leaves the conveyor belt. But with tactile feedback, the robot can slide its fingers along the fabric, feeling the folds, and then compress it into a uniform stack. Luna: I remember seeing demos of robot arms folding t-shirts a few years ago, but they always looked very staged. Lucas: They were. The lighting was perfect, and the shirts were pre-positioned. Today's versions use tactile data to correct for misalignment in real time. If the shirt slides two centimeters to the left during the fold, the robot feels the tension change in the fabric and re-centers its grip before completing the maneuver. It is essentially touching its way through the task rather than guessing based on pixels. Luna: Does this mean we are getting closer to general-purpose home robots? Lucas: We are, but the timeline is still measured in years, not months. The complexity increases exponentially when you move from a controlled factory floor to a messy living room. However, the foundational technology being built for industrial tactile sensing is directly transferable to consumer applications. Startups like Agility Robotics and Tesla are watching this space closely because the ability to interact safely with soft objects is the primary safety requirement for humanoid robots working near humans. Luna: Safety is definitely a huge driver for adoption in consumer spaces. Lucas: Absolutely. If a robot arm crashes into a human hand, you want it to feel the contact and retract immediately. Tactile sensors provide that immediate collision detection layer that visual systems alone cannot match due to latency issues. A camera frame might update thirty times a second, but a tactile sensor updates thousands of times a second. That speed difference is critical for physical safety. Luna: So speed of data matters as much as accuracy in this context. Lucas: It does. And this brings us to another interesting trend. We are seeing a convergence of vision and touch in the software layer. Engineers are training neural networks to process both visual input and tactile data simultaneously. This multimodal learning allows the robot to build a more robust internal model of the world. It doesn't just see the object; it understands its material properties based on how it interacts with the object. Luna: That sounds like a much more human-like way of perceiving reality. Lucas: It is. Humans rarely rely on sight alone when manipulating objects. We close our eyes sometimes to focus entirely on touch. Robots are starting to do the digital equivalent. By weighting tactile data heavily during manipulation tasks, the system ignores visual noise or glare that might confuse a pure vision model. This makes the robots more reliable in poorly lit or chaotic environments. Luna: Are there any downsides to adding this level of sensory complexity? Lucas: There are maintenance challenges. Sensors get dirty, damaged, or wear out. A tactile fingertip is a mechanical interface. It needs to be durable enough to withstand millions of cycles. Companies are currently testing different materials for sensor covers, looking for a balance between sensitivity and longevity. Some are using silicone coatings, others are exploring graphene-based films that are thinner and potentially more durable. Luna: Material science is playing a bigger role here than I expected. Lucas: It really is. You cannot separate the software advances from the hardware constraints. The best algorithm in the world won't work if the sensor breaks after ten thousand uses. So we are seeing a lot of investment in ruggedizing these components. The goal is to make tactile sensors as cheap and durable as the motors themselves. Luna: That seems like the key to mass adoption across all industries. Lucas: Without a doubt. Once the hardware cost drops below a certain threshold, which looks likely by late twenty twenty seven, we will see tactile-enabled robots in agriculture, healthcare, and retail. Imagine a robot stocking shelves who can check if a bottle is cracked before placing it, or a medical assistant who can gauge the pressure of a bandage. The applications are vast. Luna: It feels like we are entering a new phase of hardware intelligence. Lucas: I think so. For a long time, the narrative was all about AI brains in dumb bodies. Now, the body is becoming smart. The integration of touch is closing the loop between perception and action in a way that visual-only systems never could. It is a subtle but fundamental shift in how we design autonomous systems. Luna: Do you think this will change the job market for warehouse workers faster than anticipated? Lucas: It will accelerate the transition away from repetitive picking tasks. But it also creates demand for technicians who can maintain these complex sensor arrays. The jobs are shifting from manual labor to technical oversight. Workers will be managing fleets of robots that require more sophisticated care and calibration than previous generations of automation. Luna: So upskilling becomes just as important as the technology itself. Lucas: Exactly. The hardware is only half the story. The workforce needs to adapt to work alongside these more sensitive and capable machines. As we head into the third quarter of twenty twenty six, expect to see more training programs focused on robotic maintenance and sensor diagnostics. Luna: It is a reminder that technology changes the nature of work, not just the volume of it. Lucas: Right. And as these robots become more dexterous, the definition of 'work' they can perform expands. We are no longer talking about lifting heavy boxes. We are talking about delicate manipulation tasks that require a level of finesse previously reserved for humans. That opens up entire new categories of automation. Luna: What is the one thing you think investors should watch in this sector over the next year? Lucas: Watch the companies that are vertical integrating their sensor production. Those who can manufacture their own tactile arrays in-house will have a margin advantage and faster iteration cycles. Standalone sensor suppliers might struggle against giants who can bundle touch capabilities directly into their end-effectors. Luna: Control over the core component seems to be the winning strategy. Lucas: It usually is in hardware. Just look at the semiconductor industry. The company that controls the chip controls the pace of innovation. In robotics, the sensor is the new chip. Whoever masters the low-cost, high-density tactile array will define the next decade of automation. Luna: A compelling view on where the industry is heading. Lucas: It is. And as we wrap up, I wonder if you think the public will ever notice the difference between a robot that sees and one that feels, or if it will remain an invisible backend improvement. Luna: I think users will notice the lack of broken items and the smoothness of interactions long before they understand the technology behind it. Lucas: True. Reliability speaks louder than specs. But for those of us who build these systems, the tactile revolution is just beginning. The question now is whether we can scale this intelligence without scaling the cost. Luna: That remains the ultimate hurdle for widespread global adoption. Lucas: Let's see if the supply chain holds up through the rest of the year. It will be an interesting test of engineering versus economics.