Snapbot can have up to six legs at a time. There are three types of magnetically coupled removable legs. A locomotion algorithm determines the robot’s configuration in real-time and chooses a combination of motions to propel the robot on a forward path. (Credit: Disney Research)

Scientists at Disney Research, Pittsburgh, PA have developed a modular, reconfigurable legged robot named Snapbot that can move forward, interact with its environment, and perform other tasks based on a number of possible configurations. This system identifies its current configuration using only internal sensors and utilizes a corresponding motion strategy to complete its task. The motion strategy changes as it is physically reconfigured in real-time.

A central component of Snapbot is the 3D-printed base unit, or body, which houses a controller and battery for untethered operation. The body features distributed electrical-mechanical connectors that locate, secure, power, and communicate with the modular legs. Mating connectors couple magnetically. An array of magnets restricts or allows certain orientations of leg attachment. In the center of this coupling is an 8-pin spring-loaded electrical connector that connects power throughout the system and transmits data between the various actuators, sensors, and the controller.

The motion controller determines the location and identity of attached components, which is referred to as the configuration, in real time by pinging all possible actuators and peripherals associated with the system. Based on the determined configuration, a combination of motions is executed to propel Snapbot on a path forward.

The locomotion algorithm is implemented on the controller board of the robot’s body. The algorithm enables Snapbot to move in various configurations with one to six legs by recognizing configuration changes and selecting the best locomotion method. The algorithm is composed of two parts — recognition of Snapbot’s current configuration and its motion control. In the researchers’ first implementation, the goal task of Snapbot was limited to traveling straight forward.

Snapbot’s three kinds of 2-DOF or 3-DOF legs all have various configurations. As the result, the robot can have 700 different configurations. The researchers are planning to make Snapbot learn how to move using reinforcement learning or evolutionary algorithms. For this, other sensors including a camera will be attached either to Snapbot or externally.