RunBot
RunBot is a miniature bipedal robot which belongs to the class of limit cycle walkers. Instead of using a central pattern generator it uses reflexes which generate the gait. The reflexes are triggered by ground contact sensors in the feet which then activate the motors. The generation of the walking gait is straightforward: when a foot touches the ground the other leg is lifted upwards so that the robot falls forward. This then causes this leg to touch the ground and so forth. The walking speed can be improved by means of reinforcement learning because there are only a few parameters in this scheme. RunBot was built in 2005 by Tao Geng as part of his PhD under supervision of Prof Woergoetter and after an idea by Dr Porr to use a walking robot to benchmark reflex based reinforcement learning rules. Its movements and adaptability are based on the work of neurophysiologist Nikolai Bernstein. Since its inception the RunBot has undergone numerous design iterations, for example where a moveable upper body mass on the robot keeps the walking pattern stable even on uneven terrain.
RunBot [1][2] is a miniature bipedal robot which belongs to the class of limit cycle walkers. Instead of using a central pattern generator it uses reflexes which generate the gait. The reflexes are triggered by ground contact sensors in the feet which then activate the motors. The generation of the walking gait is straightforward: when a foot touches the ground the other leg is lifted upwards so that the robot falls forward. This then causes this leg to touch the ground and so forth. The walking speed can be improved by means of reinforcement learning because there are only a few parameters in this scheme. RunBot was built in 2005 by Tao Geng as part of his PhD under supervision of Prof Woergoetter and after an idea by Dr Porr to use a walking robot to benchmark reflex based reinforcement learning rules. Its movements and adaptability are based on the work of neurophysiologist Nikolai Bernstein.[3]
Since its inception the RunBot has undergone numerous design iterations,[4][5] for example where a moveable upper body mass on the robot keeps the walking pattern stable even on uneven terrain.
Design
[edit]The locomotion system is kept simple with four motors: one on each of two knees, one on each of two hips. The sensory system is of similar simplicity, with the ability to detect the ground contact and the angles of the hips/knee motors. The motors are controlled by force and not by angle.
References
[edit]- ^ Geng, T.; Porr, B.; Wörgötter, F. (2006) A Reflexive Neural Network for Dynamic Biped Walking Control. Neural Computation, May 2006, Vol. 18, No. 5, Pages 1156-1196 Posted Online March 13, 2006. {{doi:10.1162/neco.2006.18.5.1156}}
- ^ T. Geng; B. Porr; and F. Wörgötter (2006) Fast Biped Walking with A Sensor-driven Neuronal Controller and Real-time Online Learning. International Journal of Robotics Research, 25(3), p 243-259, March 2006, Sage press.
- ^ "Record speed set by walking robot". BBC News Online. 2006-05-04. Retrieved 2007-07-14.
- ^ Geng, T.; Porr, B.; Woergoetter, F. (2006) Fast biped walking with a reflexive controller and real-time policy searching. Advances in Neural Information Processing Systems 18. MIT Press, Cambridge, MA
- ^ Manoonpong, P.; Geng, T.; Kulvicius, T.; Bernd Porr; Woergoetter, F. (2007). Adaptive, Fast Walking in a Biped Robot under Neuronal Control and Learning. PLoS (Public Library of Science) Computational Biology (PLoS Comput Biol), 3(7), e134. doi:10.1371/journal.pcbi.0030134
External links
[edit]Runbot's creators recorded demonstrations of RunBot:
- "walking at a steady speed". The International Journal of Robotics Research. Retrieved 2007-07-14.
- "learning to accelerate". The International Journal of Robotics Research. Retrieved 2007-07-14.
- "robot keeps walking with multi-level control". New Scientist. Archived from the original on 2021-12-21. Retrieved 2007-07-14.
- "Bernd Porr's RunBot page".
- "Poramate Manoonpong's RunBot page".