- 著者
-
BERTHOUZE Luc
- 出版者
- 独立行政法人産業技術総合研究所
- 雑誌
- 若手研究(B)
- 巻号頁・発行日
- 2003
The purpose of the study was to understand the mechanisms underlying early locomotor skill acquisition by replicating the bouncing study of Goldfield et al. (developmental psychology) by using a robot strapped to a Jolly Jumper. Such replication was expected to help construct hypotheses and predictions on the key requirements for the acquisition of motor skills in humanoid. A small biped humanoid robot was constructed for the purpose of the experiment. In human infants, the natural compliance of the infant's musculoskeletal system reduces the dynamic loads of bouncing. In robots, however, mechanical compliance has a negative influence on positional accuracy, stability and control bandwidth. Compliant extensions were constructed using visco-elastic material placed in brass bushes and mounted in series with the actuators. A compliant foot system was implemented as using springy toes and a rigid heel. While compliance provided the damping necessary to cut off oscillations at an early stage, it also induced backlash, which from a control point of view, results in delay in the feedback loop. For robust jumping performance to occur, those delays must be compensated for by the control structure. A control structure based on biologically-plausible oscillators (Bonhoeffer-Van de Pol) used as pattern generators, was developed. It was shown that the architecture displayed flexible phase locking whereby the oscillators could entrain to sensory feedback from sensors placed under the feet, even in the presence of large delays. This property of flexible phase locking makes the choice of a particular organization of oscillators less critical than when harmonic oscillators are used, especially when self-tuning of the oscillators' time constants is possible ("tuning" phase studied in year 15). Robustness to environmental perturbations was tested systematically. The control framework showed to be very flexible, with rapid adaptations to changes in ground height for example.