Beanbag Robotics

This page deals with research I pursued 2007 at my research visitation at Cornell University, Ithaca, USA. It presents robots designed in a very minimalistic way, that are even unable to steer at the individual level, but show full navigational capabilities as a swarm.


In recent years, growing research interest in the field of swarm intelligence inspired projects showing a variety of behavioral patterns in robotic swarms. In most experiments, individual swarmers possess full navigational capabilities like the ability to move in fore- and backward direction as well as directional steering.

Problems and Methods

Such capabilities place severe constraints in the size and costs of each individual unit and consequently limit the number of individual swarmers.

In this work, it is showed that similar complex swarm behavior can be elicited from individuals that possess much less degrees of control freedom. While the named tradional swarmers are capable of directional steering and forward and backward velocity control, our swarmer models possess fewer navigational capabilities: They only are capable to control the velocity of a noisy forward locomotion.

This behavior can be implemented in roboticy by agents performing simple forms of vibration. However, this causes problems. If swarmer, for example, search for a light source, they can speed up in the right points of time, but gradually get lost in a potential unlimited environment (swarmer trajectories of such behaviore are provided in the image on the left. The light source is yellow, the starting a area of the swarmers is blue).

The interaction of those simple units encosed in a passive membrane (see image below) in the context of artificial evolution then causes a moveability of the whole swarm impossible at the individual level. We understand navigation as a qualitative new capability, which emerges at the collective level rather than being possessed by each single individal.

Beanbag Roboter

As experiments, we chose obstacle avoidance, reaching light sources through narrow gaps and food search and delivery at a nest.


This work contributes both to the fields of robotics and swarm intelligence. It contributes to robotics by providing a new paradigm to construct highly redundant, scalable, liquid-like robots out of such simple swarmers (image on the right). It contributes to swarm intelligence by showing how full navigational capabilities can be evolved out of a swarm of units not even able to steer.