What happens if a sidewalk robot crosses the path of a car that can’t turn out of the way?
The video from Miami has the answer to this question in a chilling way: a small food delivery robot is standing on the railroad tracks as a passenger train approaches and crashes into it, tearing it apart. The video was taken by Guillermo Dapelo, who said that he noticed the robot around 8 p.m. while walking his dogs and that it seemed stranded for about 15 minutes before the accident.
In the video, the voice of Dapelo is heard repeating a phrase that has since become the caption of the incident in several reposts: “Oh, it’s going to crash it,” he says moments before the train reaches the crossing. Dapelo also described an improvised human-in-the-loop moment an Uber Eats delivery driver reportedly contacted Coco Robotics, the operator of the robot, “to let them know where it was standing and within a matter of minutes the train approached and everything else is captured on the video.”
As Coco Robotics pointed out, the problem was a hardware issue that was “extremely rare” and happened when the robot was not directly involved in the delivery of the package. Coco Robotics has also tried to point out the safety plan that is part of the rollout of the robots: “Safety is always our top priority, which is why our robots operate at pedestrian speeds, yield to pedestrians, and are continuously monitored in real-time by human safety pilots,” Coco’s vice president and head of government relations Carl Hansen said in a statement. He also said that the company had been operating in Miami for over a year and had completed the same routes several times a day without any issues.
The underlying engineering principle here is not that a robot has failed but that there are edge cases where urban infrastructure does not behave like a sidewalk. Delivery robots are designed for slow, pedestrian-scale problem-solving: route calculation that takes into account a blocked curb cut, sensors that understand a crosswalk light, and human operators who can intervene when something unexpected comes into view. Many of these systems use cameras and other sensors to build a real-time model of the world and then make a decision about whether to slow down, yield, or reroute. This toolbox is exactly what is needed for busy city blocks, where most hazards can be avoided. A rail crossing is not. When a robot is stuck on the tracks, the problem shifts from navigation to clearance, and time becomes an issue.
This is why a stuck robot is more than a failed device. The same urban environments that are testing the automation of delivery are also incorporating robots into a regulatory framework that varies by state and local jurisdiction. At least 23 states had laws on the books about sidewalk delivery robots by the end of 2022, but the details can vary wildly, from weight restrictions to speed restrictions to where robots are allowed to go. These details define the technology that will be used, and they also define the “escape hatches” that a designer can create: how quickly a robot can determine that it is stuck, what moves a remote control operator can make, and what kinds of environments a robot is allowed to operate in to begin with.
As delivery robots move from campus curiosities to neighborhood mainstays, the Miami incident proves a simple principle: the more predictable the environment, the easier it is to achieve autonomy, and the less predictable, the more rigid the environment, the harder it is to achieve autonomy. Tracks are rigid. Trains are, too.
