Gorillas in the cyber-mist

Roboticists in Japan have designed a robot gibbon that swings and learns just like the real thing – almost

Most of us have, at one time or another, been swept away by the languid grace and gravity defying comedy of a gibbon in full swing. But now, incredibly, scientists at Japan’s Nagoya University have replicated this behaviour in a robot that uses genetic algorithms to teach itself to swing from branch to branch like a real long-armed ape.

Brachiator III, at the heart of the university’s research into autonomous robots; uses a computer, 14 motors and a fully articulated body to teach itself to swing.

Based as closely as possible on the physiology of a real gibbon, this 13-link, 12 jointed robot is able to carry out extremely dextrous moves. The 14 DC motors, that include two for the robot’s grippers, drive each of the joints.

Actuator problems mean the robot is about 1Kg heavier than its tree dwelling cousins, but this is superficially the biggest difference.

A surprising amount of robotic research has gone into the motion of gibbons, but this is the first time, say the Japanese, that such a robot has been able to move in three dimensional space. Previous efforts have been limited to movement in a two-dimensional plain.

Moving dynamically from branch to branch and swinging its body like a pendulum, the robot’s hierarchical controller divides its behaviour into, first a swinging action which stores energy prior to the transfer (preliminary swing mode), and then a locomotion action which is the actual transfer operation (locomotion mode). These actions are then divided further into the following fundamental behaviours: leg swing, body rotation, leg stretch, body rotation, body lift and arm reaching.

Spatial awareness is provided by a Quick MAG System III real-time tracking system that measures the three-dimensional locations at eight measuring points on its body using two CCD cameras.