Robotics & Automation
The Robot That Learned to Be Soft
The hardest problem in robotics isn't making machines stronger — it's teaching them to be gentle enough not to crush a grape.
The Idea
For most of robotics history, the field has been obsessed with rigidity. Steel joints, servo motors, precise tolerances — the assumption was that control required stiffness. If you could specify every degree of movement exactly, you could make a robot do anything. The problem is that the real world isn't a factory floor. It's full of objects that squish, surfaces that yield, and situations where the right response to unexpected resistance is to back off, not push harder. Soft robotics starts from a completely different premise: what if compliance — the ability to deform, flex, and absorb — is a feature, not a bug? Instead of rigid frames and motors, soft robots are built from silicones, hydrogels, pneumatic chambers, and materials that borrow more from biology than engineering. A soft gripper doesn't need to know the exact dimensions of what it's picking up; it conforms around the object the way a hand does. The deeper insight is that soft systems offload computation to physics. A rigid robot needs algorithms to calculate how to grasp something without crushing it. A soft one solves that problem partly through its own material properties — the intelligence is embedded in the structure itself. Researchers call this 'morphological computation', and it's a genuinely different philosophy of what it means for a machine to be smart. The body, not just the brain, is doing the thinking.
In the World
In 2016, a team at Harvard's Wyss Institute unveiled an octopus-inspired soft robot so pliant it could squeeze through a gap smaller than its own body — something no rigid robot could attempt without catastrophic failure. But the more commercially consequential example came from a startup called Soft Robotics Inc., whose pneumatic grippers began appearing on food production lines handling everything from muffins to cherry tomatoes. The problem they were solving was deceptively mundane: existing robotic arms kept bruising produce. Fruit, it turns out, is terrible at tolerating the kind of repeatable, forceful grip that works fine on a car chassis. More recently, researchers at MIT and elsewhere have been developing soft robots for medical use — devices small and flexible enough to navigate the coils of the intestine or the branching passages of the lung without tearing tissue. One prototype, published in 2023, used magnetically steered soft tendrils to move through the body guided by an external field, the robot's flexibility meaning it could bend around corners that would snap a conventional catheter. Perhaps the most striking demonstration of the philosophy came from a DARPA-funded project that produced a soft robot capable of crawling across uneven terrain by inflating and deflating internal chambers in sequence — no wheels, no legs, just rhythmic deformation. It looked less like a machine and more like something you might find under a rock at low tide.
Why It Matters
Soft robotics isn't just a niche engineering curiosity — it's quietly rewriting the assumptions about where robots can go and what they can do. The places humans most want robotic help tend to be exactly the places rigid machines fail: cluttered homes, hospital wards, agricultural fields, the inside of a human body. All of these environments reward adaptability over precision, and compliance over force. There's also something philosophically interesting happening here. The field is drawing heavily from biology — cephalopods, plant tendrils, and muscular hydrostats like the tongue are recurring inspirations. In trying to make robots useful in messy human contexts, engineers are essentially reverse-engineering what evolution figured out hundreds of millions of years ago: that softness is a survival strategy. For anyone thinking about the future of automation, the takeaway is that the robots most likely to enter intimate spaces — your kitchen, your hospital room, your ageing parent's home — probably won't look like the metallic humanoids of science fiction. They'll be quieter, stranger, and considerably squishier than that.
A Question to Ponder
If embedding intelligence into a material's physical structure rather than its software is genuinely possible, what else might we currently be solving with computation that could instead be solved with better-designed matter?
Get a new one of these every morning.
Start learning with Thinkable