Loading prices…
STKR NewsSTKR News0 of 3 free this month
Future Tech

Video Friday: Drone Squeezes Itself to Fit Through Tight Spaces

New robotics research highlights a shift from rigid machines to adaptive, foldable, and low-cost systems that prioritize real-world utility over lab-perfect aesthetics.

Originally on IEEE Robotics →
AB

Adrian Boysel

Contributor

Nov 20, 2026

4 min read

Photo illustration / STKR News

I spent the morning looking at the latest batch of robotics breakthroughs from IEEE Spectrum, and if you filter out the slick editing, there is a clear trend emerging for those of us building in the physical AI space. We are moving away from the era of "perfectly rigid" machines. For years, the goal was precision through stiffness. Now, the industry is leaning into morphing, folding, and modularity. It is less about making a robot that can overcome any obstacle and more about making a robot that can change its own shape to fit the world as it actually exists.

The End of the Rigid Quadrotor

One of the most practical developments I saw involves a quadrotor that literally squeezes itself through tight spaces. Usually, if a drone hits a wall or tries to fly through a gap narrower than its frame, it crashes. The physics of flight traditionally demand a fixed geometry to maintain stability. But researchers at the Autonomous Robots Lab have developed a design where the side arms bend on contact. It can shrink from a 31.3 cm width down to 14.1 cm while staying in the air.

For founders in the inspection or search-and-rescue niche, this is a big deal. We’ve spent millions trying to build better collision avoidance sensors. This research suggests that maybe the hardware should just be more forgiving. If your robot can survive a bump and keep flying by folding its arms, you don't need a thousand-dollar sensor array to avoid every single twig. It’s a hardware solution to a software problem, and those are often more reliable in the field.

Modular Soft Robotics and the "Body on Demand"

Then there is the TeCoBot system. These are snake-like soft robots that can link up to form larger structures. A single unit can crawl or roll, but when they chain together, they can lift heavy objects or climb complex pipe systems. What I like here is the lack of specialized docking hardware. The same claw they use to grab objects is the one they use to grab each other.

If you are building in the industrial automation space, take note. We usually think of robots as single-purpose tools—a robotic arm, a rover, a drone. The TeCoBot approach suggests a future where the "robot" is just a collection of nodes that rearrange themselves based on the task. It’s messy, but it’s efficient. It reduces the number of spare parts you need to carry and increases the versatility of the fleet.

The $650 Dexterous Hand Problem

We need to talk about the PHD-Hand. Most dexterous robotic hands are a nightmare to maintain because they rely on tendons. If you’ve ever worked with cable-driven systems, you know they stretch, snap, and requires constant re-tensioning. It’s fine for a lab, but it’s a non-starter for production manufacturing.

Neel’s work on a 3D-printed hand using rigid linkages for under $650 is the kind of "builder-first" engineering I love. It achieves human-like dexterity without the fragility of tendons. For those trying to bring AI into the physical world, the cost of the "end effector" (the hand) has always been a barrier. If we can get the bill of materials down to triple digits while keeping the hardware durable, the barrier to entry for general-purpose robots drops significantly.

Physical AI in Production

Ben Burchfiel from Walden Robotics recently gave a talk at CMU that resonates with my skepticism about "lab-only" AI. Walden deployed their first robot into a factory just two months after starting the company. That is incredibly fast. Their core thesis is that you cannot build physical AI in a simulation alone; it has to learn on the job, in a real factory, doing real work.

This is a wake-up call for founders who are stuck in the "research phase." The data you get from a robot failing in a real production line is worth ten times the data from a successful run in a simulated environment. The "inflection point" Burchfiel mentions isn't just about better LLMs; it's about the feedback loop between hardware and the real world getting shorter.

Autonomy vs. Human Skill

Finally, the documentary on the SWIFT racing drone from the University of Zurich is worth a watch. We saw AI beat world-champion human pilots back in 2022, and the tech has only matured since. But notice the distinction: the AI isn't just "faster" at processing; it's better at navigating the physics of the environment in ways a human brain isn't wired for.

However, we shouldn't get distracted by the spectacle. Racing is a controlled environment. The real challenge for builders is taking that autonomous speed and applying it to unpredictable environments—like a construction site or a home. Speed is solved. Adaptability is the next frontier.

What This Means for Builders

  • Prioritize Durability Over Precision: The folding drone and the linkage-based hand show that being "tough" and "flexible" is more valuable than being perfectly calibrated.
  • Lower the Barrier to Failure: If your robot costs $600 to build instead of $60,000, you can afford to let it fail, learn, and iterate in the real world.
  • Modular Thinking: Stop trying to build the one perfect robot. Build components that can work together to solve multiple problems.

The takeaway here is simple: stop waiting for the perfect sensor or the perfect simulation. The future of robotics is being built by people who are making cheaper, folding, clicking, and sometimes clunky machines that can survive the mess of reality. If your robot can't take a hit or change its shape, it’s probably not going to make it out of the lab.


Read the original at IEEE Robotics →

The Brief

Stay Updated on Cutting-Edge Tech

A six-minute morning dispatch on the markets and the technology shaping them.

Free. No spam. Unsubscribe anytime.

Write for STKR

Become a Contributor

Earn $STKR for published stories on markets, protocols, and culture.

  • Earn $STKR for every published piece
  • Editorial support from the STKR desk
  • Byline visibility across the network
  • First look at the upcoming creator program
Apply to Write

Keep reading

All stories

Comments

24 reader responses