...go to home

Self-Exploration of the Stumpy Robot with Predictive Information Maximization

       by Georg Martius, Luisa Jahn, Helmut Hauser, and Verena V. Hafner,
       see here for the reference and here for the pdf
Abstract: One of the long-term goals of artificial life research is to create autonomous, self-motivated, and intelligent animats. We study an intrinsic motivation system for behavioral self-exploration based on the maximization of the predictive information using the Stumpy robot, which is the first evaluation of the algorithm on a real robot. The control is organized in a closed-loop fashion with a reactive controller that is subject to fast synaptic dynamics. Even though the available sensors of the robot produce very noisy and peaky signals, the self-exploration algorithm was successful and various emerging behaviors were observed.

Video S1: Different locomotion behaviors. The robot starts with low movements. After it is relocated manually it locomtes with high amplitude. Straight and curved locomotion is mixed.

Video S2: Turning at the spot. The robot performs several turns. At time 2:10 the robot tipped over and was then hold by hand to illustrate that it does not move without the environmental feedback (at least not without adaptation to the new situation). Later in the video the robot explores different movements and falls a lot.

Video S3: Low amplitude, behavior resulting in fast locomotion.

Video S4: Swinging behavior. The robot swings to either side and moves the top part to the center short before it tips over. This results in a low frequency oscillation with high amplitude.

Video S5: Stumpy in the wok (Chinese cooking pot) The wok is loaded with books in order to make the system more stable. The video shows the robot in the rotational mode. The robot is stopped manually at time 0:26 and the controller excites the rotation mode again.

Video S6: Stumpy in the wok (Chinese cooking pot) II The wok is now loaded with a solid weight. The video starts from the beginning and shows how the controller starts to amplify the spontaneously excited rocking motion. At time 1:00 the robot falls over. After lifting manually it continues to revolve.

This document was translated from LATEX by HEVEA.