The underlying mechanism of taxis locomotion by the true slime mold involves local sensory feedback of the bio-chemical oscillators. This feedback can be extremely simple: when a local oscillator is pushed strongly by the protoplasm under high pressure, it is stretched even if it is trying to contract. Based on this biological finding, I designed a mathematical model of the local sensory feedback, and embedded it into a slime mold robot (both the simulation model and real robot on the YouTube videos bellow) as an autonomous decentralized control system. There are two unique features of this robot:
Simulation and experimental results show that this robot exhibits truly supple and adaptive locomotion without relying on any hierarchical structure. The results obtained indicated that the autonomous decentralized control is a promising approach for soft-bodied robotic systems. This control scheme has also been applied to snake-like and quadruped robots by my colleagues in the Ishiguro laboratory (Owaki et al. Interface 2012, Sato et al. Bioinspir. Biomim. 2011).
真性粘菌変形体に着想を得たソフトな身体を有するアメーバロボット [ICRA ’10などで発表]
生き物は，脳や神経を持たない単純な有機体でさえも，実時間かつ合目的的に振る舞う適応的な知を有しています．例えばアメーバのような，脳もなければ神経もない単細胞でも，外部環境の潜在的危険—例えば，高熱，低温，とがった針，多くの振動—から，即座に泳いで逃げることが出来ます．この脳のない生き物の中の振る舞いには，われわれ人間を含めすべての生物に普遍的な生存脳機能とも言える情報処理系を有しています ．そこで，本研究では非常に巨大な単細胞生物である真正粘菌変形体に着目し，脳も神経も介さない生物の逃避行動や探索活動の実現を試みました．われわれの開発したロボットでは，下記のムービーに示すように走性（刺激に向かって進む行動）の実現を自律分散制御則のみで実現しました．最近の研究では，環境探索運動と走性運動の実現を自律分散制御則により明示的な制御則の切り替えなしに実現しました．
下記に紹介するロボットの共通点は，３つです：
・体積を一定に保つ機械的な拘束がある（風船やシリンダ内の空気など）
・身体部位間に配置された柔らかいアクチュエータが，その体積を押し合い圧し合いする．
・各アクチュエータは，柔らかい材料にかかる応力が小さくなるように，伸縮運動を周期的に繰り返す．
このような単純なルールがあるだけですが，多数の要素が機械的に相互作用することにより，走性という機能を実現すること出来ます．
[Artificial life 2013, Bioinspir. Biomim 2013, Adaptive behavior 2015a]
Another advantage of soft robots is that versatile behaviors can be generated using the large degrees of freedom in their bodies. This is one of the fundamental strategies for animals adapting to unexpected situations. In order to investigate the capability of the proposed controller, I simplified the true slime mold robot into hydrostatically coupled oscillators (consisting of passive actuators, air cylinders, and tubes to connect them). Despite this simplicity, the real physical robot produced versatile oscillatory patterns and spontaneous transitions among the patterns by exploiting the mechanical (hydrostatic) interplay (the simulation model was published in Artificial life 2013, and the real physical robot and further analysis was published in Bioinspir. Biomim 2013). Based on the oscillator system, we built a modular robot with local stiffness changes that depended on the presence of an attractant; the robot was able to switch from exploratory to taxis locomotion (Adaptive behavior 2015a).
These results also indicate that mechanical interactions can transfer information between oscillators (distributed controller) without a designing a specific communication process between them. This interaction between local controllers is unique and sharply contrasts with many proposed CPG-based controllers using ‘well-designed networks’. I also believe that studying and reproducing these behaviors of the coupled oscillators can contribute to understanding more universal motion control of animals, such as biped and quadruped locomotion, because coupled oscillator systems and rhythmic motion are ubiquitous in all living systems.
In real living systems adaptive mechanisms with different time constants can co-exist without causing conflicts in the body (e.g., reflex, learning, growth and evolution). This enables living systems to survive in the face of overwhelming environmental changes, which a single adaptation mechanism would not allow. In the true slime mold at least two adaptive mechanisms exist: one is a contraction mechanism that generates cyclic oscillations with a period of 1-2 minutes and another is a morphological change producing and eliminating pseudopods over a time-scale 10 times longer. Inspired by this, we have designed a mathematical model and real physical robot incorporating these two mechanisms as decentralized controllers. Numerical and experimental results show that by combining the controllers with different time constants, a robot can use the proposed model to successfully negotiate a narrow aisle by deforming its body shape dynamically.