The Gap Between a Robot Demo and a Robot That Actually Works
A co-founder of agricultural robotics firm Burro explains why the hardest part of building autonomous machines is not the lab work. It is Tuesday morning in the rain.
Key points
- Burro, a Philadelphia-based agricultural robotics company, ran its first field demo in 2018 and spent years learning the gap between that demo and real-world deployment.
- Outdoor robots face temperature swings from below freezing to 120 degrees Fahrenheit, changing light conditions, and terrain that shifts daily.
- Burro co-founder Vibhor Sood says the missing research area is infrastructure-free localisation: letting a robot know exactly where it is and what surrounds it, without GPS or fixed reference points.
- Every real-world failure a Burro robot experiences in one location feeds back into the system and makes every other unit in the fleet more reliable.
A robot that works in a demo and a robot that works on a rainy November morning are not the same machine. Burro, an agricultural robotics firm based in Philadelphia, learned that the hard way, and co-founder Vibhor Sood has set out exactly why in a piece first reported by The Robot Report.
Burro builds autonomous robots for outdoor farm settings, machines that can carry loads, mow, and patrol fields without any fixed infrastructure like GPS towers or controlled lighting. The company did its first demo in 2018. It went well. Then the real work began.
The problem is what Sood calls the demo trap. A demonstration is built for a best-case scenario: controlled conditions, a prepared audience, a short time window. Real deployment means flat light, wet ground, workers approaching from unexpected angles, and operations in locations the robot has never visited. Most robotics companies, Sood argues, spend too long optimising for the demo while the harder deployment problem sits unsolved.
Burro had no luxury of waiting. Farm environments do not bend to a machine's preferences. Agricultural workers were not going to change how they moved through fields to help a robot cope. The company had to confront real-world failure early and learn from it fast.
Why does outdoor robotics fail where indoor robotics succeeds?
Indoor robots operate in controlled spaces: fixed lighting, flat floors, predictable movement. A warehouse does not change shape overnight. A construction site does. A port yard runs around the clock, mixing heavy vehicles with unpredictable human traffic. A farm row looks completely different at dawn versus dusk, in summer versus winter, in dust versus mud.
Sood's core argument is that no amount of simulation, however detailed, solves this. You only solve outdoor robotics by being outdoors, accumulating real operational data, failing safely, and iterating quickly. A failure in one field, once absorbed and corrected, makes every robot in the fleet smarter.
The specific technical gap Sood identifies is infrastructure-free localisation and perception. In plain terms: a robot needs to know precisely where it is and what surrounds it, even as its sensors wear down and the environment shifts around it, all without any supporting infrastructure. Indoor robotics received serious investment in solving this problem about a decade ago. Outdoor, unstructured environments have not received the same attention. That gap, Sood says, is the primary remaining bottleneck.
The lesson for anyone watching autonomous systems enter their workplace or neighbourhood is a practical one. A system that impresses in a controlled setting is not yet a system you can depend on. Dependency comes from field hours, not lab hours. Burro says it now holds eight years of that kind of data, earned in full.
When a robot does earn that dependency, users reorganise their entire workflow around it. At that point, a failure feels less like a broken gadget and more like a power cut. The reliability bar set by real users is higher than any lab test will prepare a company for.
