Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Industrial Robots Struggle With Soft Packaging
Transcript
- Lucas: Look at a warehouse floor today, specifically one that’s been fully automated over the last three years. You see arms moving at impossible speeds, lifting pallets of canned goods, sorting boxes with laser precision. It looks like sci-fi made real. But if you walk to the end of that line, where the final product gets packed into those crinkly, soft plastic bags of chips or frozen vegetables, you often find a human standing there. Luna: Right, because the robot just can't tell the difference between a full bag and an empty one, or it crushes the contents trying to grip it. Lucas: Exactly. And here is the thing most people miss about this industry right now. We are not facing a shortage of robotic arms. We are facing a shortage of dexterity for non-rigid objects. The hardware exists, the software is getting smarter, but the fundamental physics of soft, deformable materials breaks almost every rule we use for hard-body automation. I want to talk about why this specific gap is becoming the biggest bottleneck in logistics efficiency in September of twenty twenty-six. Luna: It feels like we solved the navigation problem with vacuums and delivery bots, so why is picking up a bag of flour still so hard? Lucas: Because navigation is a map problem, and picking up a bag is a geometry problem that changes every millisecond. Let me give you a concrete example from a mid-sized food distribution center I looked at recently. They tried to automate their packing line for frozen meals. The meals come in flexible foil pouches. The robot was supposed to pick them from a conveyor and stack them into a box. Within four hours, the system had jammed twice and crushed three hundred packages. Luna: So the vision system couldn't lock onto a consistent shape because the bag shifts every time it moves? Lucas: That is part of it, but it goes deeper. The camera sees a blob. It calculates a center point based on rigid body assumptions. When the arm reaches down, gravity pulls the top of the bag down, the sides bulge out, and the center of mass shifts before the gripper even touches it. By the time the fingers close, they are squeezing air, not product. Or worse, they squeeze too hard and rupture the seal. It is a feedback loop that standard PID controllers just cannot handle fast enough. Luna: It sounds like the error margin for a rigid box is zero tolerance, but for a soft bag, the error margin keeps changing. Lucas: Precisely. A cardboard box has six faces and eight corners. A bag has infinite potential configurations. This is why we see these expensive failures. Companies spend millions on the latest LiDAR and high-resolution rgb d cameras, hoping that more data will solve the problem. But more data does not help if your model assumes the world is made of bricks. We need models that understand fluid dynamics and material science, not just pixel tracking. Luna: So the hardware needs to adapt, or the software needs to become significantly more predictive about how materials behave under stress. Lucas: Both. But let's look at the economic pressure here. Labor costs in logistics have risen steadily since twenty twenty-two. Automation was supposed to be the hedge against that. But if you install a fifty thousand dollar robotic cell and it only works eighty percent of the time because of soft packaging issues, you have actually increased your cost per unit. You now have maintenance staff, debugging engineers, and still need that human backup operator. The ROI disappears. Luna: That is a harsh reality check for investors who thought robotics was a solved problem by now. Lucas: It is. And this brings me to the donation segment, which is actually relevant to why we cover these technical nuances so deeply. We deliberately don't run ads on these shows. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Luna: Yeah, keeping the signal to noise ratio high means we can focus on the actual engineering hurdles rather than chasing clicks for sponsors. Lucas: Thanks for considering that. Now, back to the mechanics. One company, a startup called SoftWing Robotics, has taken a completely different approach. Instead of trying to make the robot see better, they changed how it feels. They developed a gripper that uses micro-vibrations. As the fingers close around the bag, the vibrations help the sensors detect the exact tension points in the fabric. Luna: Micro-vibrations? That sounds like how ultrasound imaging works, but applied to tactile feedback. Lucas: It is similar. The idea is that by feeling the resistance change in real-time, the robot can adjust its grip force dynamically. If the bag starts to slip, it tightens. If it detects the material stretching beyond a safe limit, it loosens immediately. It mimics the human proprioception we have in our own hands. We know how much pressure to apply to hold an egg without breaking it, mostly because our nervous system processes that feedback loop thousands of times a second. Luna: But robots process at milliseconds, and human nerves are biological. Is that speed gap closing? Lucas: It is. Edge computing has improved dramatically. Processing that tactile data locally on the robot arm, rather than sending it to a cloud server, reduces latency to under five milliseconds. That is fast enough to prevent crushing. However, the energy cost is higher. These active grippers consume more power than passive ones, which impacts battery life for mobile manipulators. Luna: So there is a trade-off between dexterity and operational endurance. You get better handling, but the robot needs to charge more often. Lucas: Correct. And in a warehouse running twenty-four seven, downtime for charging is money lost. So some facilities are going hybrid. They use the advanced soft-grip robots only for the delicate items, and keep simple, cheap pneumatic grippers for the heavy, rigid cases. It is not elegant, but it is profitable. Luna: Hybrid systems seem like the pragmatic bridge until the technology matures. It acknowledges that one size does not fit all in automation. Lucas: It does. And this segmentation is happening across the board. We are seeing specialized robots for pharmaceutical blister packs, specialized ones for garment folding, and specialized ones for produce. The era of the universal robot arm is stalling because universality requires compromise, and compromise fails with soft objects. Luna: So instead of one giant generalist machine, we might see a fleet of specialists, each optimized for a specific type of packaging material. Lucas: I think that is exactly where the market is heading. It reminds me of the early days of computers. Everyone wanted a machine that could do everything. Then we realized that specialized silicon, like GPUs for graphics or TPUs for AI, was far more efficient. Specialized end-effectors for robotics are the next step. Luna: It shifts the design philosophy from the arm itself to the interface with the object. The hand becomes as important as the brain. Lucas: Absolutely. And this raises a question about training data. To train these specialized grippers, companies need massive datasets of soft objects deforming under various conditions. Who owns that data? Is it proprietary to the robot manufacturer, or is it shared across the industry? Luna: That is a tricky legal landscape. If a food company generates the data on how their specific bag behaves, they might consider that a trade secret. Sharing it could help competitors optimize their packing lines. Lucas: Right. So we might see a fragmentation of standards. Each major retailer builds their own closed-loop ecosystem. Amazon has theirs, Walmart has theirs. They won't share the deformation models because that gives them an edge in speed and waste reduction. It creates a walled garden for robotics data. Luna: Which slows down innovation for smaller players who can't afford to generate that much proprietary data from scratch. Lucas: Exactly. It raises the barrier to entry. If you are a small integrator trying to sell a solution to a mid-sized bakery, you can't compete with Amazon's internal optimization tools. You have to rely on open-source models, which are less accurate. It is a classic winner-take-all dynamic playing out in hardware. Luna: So the bottleneck isn't just physics anymore. It is data ownership and market structure. The technology is constrained by business incentives. Lucas: Yes. And we should not underestimate how much business inertia plays a role. Warehouses are designed around rigid standards. Pallet sizes, box dimensions, conveyor widths. Changing the infrastructure to accommodate soft-object variability is expensive. Most companies are trying to force soft objects into rigid workflows, which is why the robots fail. Luna: They are trying to square the circle. Expecting a flexible item to behave like a rigid one in a rigid system. Lucas: That is a perfect way to put it. Until the entire supply chain, from manufacturing to shipping, adapts to the reality of soft packaging, robots will continue to struggle. It requires a systemic redesign, not just a better algorithm. Luna: So the solution might lie in packaging design itself. Creating bags that are easier for robots to handle, perhaps with reinforced edges or standardized shapes. Lucas: Some companies are experimenting with that. Rigidifying certain parts of the package without losing the flexibility of the material. It is a niche engineering challenge, but it could unlock a lot of value. Imagine a chip bag that stands up straight automatically for the robot to grab, then collapses flat for shipping. It solves the geometry problem at the source. Luna: Designing for the robot's limitation. It is interesting that we usually design for the human consumer, but now we have to design for the machine that touches it first. Lucas: We are entering a new phase of industrial design where the primary user is a sensor suite and a mechanical actuator. Human ergonomics become secondary, or at least, parallel concerns. It changes the aesthetic and functional priorities of everyday products. Luna: It makes you wonder what our daily consumption habits will look like in ten years if machines are doing the initial handling. Will packaging become more utilitarian? Lucas: Likely. We might see a split. Premium goods stay human-centric and beautifully packaged, while commodity goods become hyper-efficient and robotic-optimized. The middle ground shrinks. Luna: A bifurcation of the market based on who is holding the product. That is a stark distinction. Lucas: It is. And for now, that middle ground is where the frustration lies. The robots are good enough to be dangerous, but not good enough to be reliable. We are stuck in the uncanny valley of automation. Luna: The uncanny valley of automation. I like that phrase. It captures that sense of almost there but still failing perfectly. Lucas: And until the physics models catch up to the visual models, we will probably see more of these hiccups. It is a reminder that nature is messy, and trying to impose order on it with rigid machinery is always going to meet resistance. Luna: Unless that machinery learns to flow with the messiness instead of fighting it. Lucas: Exactly. Learning to bend, rather than break. Whether it is in materials, algorithms, or business models, flexibility might be the only true competitive advantage left.