Researchers Find Random Movement Helps Robots Navigate Crowds Faster

A team of researchers led by Harvard Ph.D. student Lucy Liu ’22 found that robots moving through crowded spaces reached their destinations faster when programmed to move with a small degree of randomness — a finding that could help researchers better understand traffic jams and pedestrian flow.

The study, published earlier this month in the Proceedings of the National Academy of Sciences, used simulations of robot swarms to test how simple movement rules affected congestion. The researchers found that when robots moved directly toward their destinations, they often became trapped in gridlock. But when they occasionally moved at random to navigate around obstacles, the group moved more efficiently.

Too much randomness, however, slowed the robots down.

The result, Liu said, suggests that researchers may be able to predict and influence crowd behavior without mapping every possible interaction among individuals — a task that has historically been difficult because of the number of paths and collisions involved.

“This research is laying the foundation for being able to tune and understand crowds of all these interacting agents,” Liu said.

Rather than treating the robots as a single coordinated system, the team modeled each robot as an individual agent following a basic set of instructions: move toward a destination, and when blocked, move randomly for a short period of time.

Liu said the simplicity of that rule was central to the study.

“If you understand what the crowd is doing on a zoomed-in level, you can also zoom out and have an idea of what will happen at the global level, and that is a hard result to get, so it’s really exciting,” she said.

The team also included Harvard School of Engineering and Applied Sciences professor Lakshminarayanan Mahadevan, SEAS robotics research fellow Justin Werfel, and Federico Toschi, an applied physics and mathematics professor at Eindhoven University of Technology in the Netherlands.

Though the study was conducted with robots, the researchers said the results could eventually inform models of human movement in crowded public spaces. Liu said future work will focus on extending the robots’ behavior to more closely resemble human motion.

Toschi, who developed the robots used in the study, said the robotic model may also help researchers separate the mechanical features of crowd movement from the psychological choices humans make in crowds.

“There are some situations where we observe in the flow of humans, we observe some things, and we are not sure if what we see is due to some mechanistic choices or psychological choices,” he said. “So if we implement with the robot some algorithm, we see that the same phenomenology emerge. Then, for example, we may rule out psychology.”

While the researchers hope to make their simulations more realistic, Liu said artificial intelligence may not be necessary. Instead, she said, adding “aspects involving more prediction” could make the models more accurate.

Werfel said the study showed that even limited decision-making can produce efficient group behavior.

“It illustrates how you can have local, decentralized rules, meaning the robots operate without a central choreographer to perform some kind of collaborative and complicated task,” Liu said.