Biosystemas

BioSystems 87 (2007) 215–223

Robot control with biological cells
Soichiro Tsudaa,∗ , Klaus-Peter Zaunerb , Yukio-Pegio Gunjia
a

Graduate School of Science and Technology, Department of Earth & Planetary Science, Kobe University, Nada, Kobe 657-8501, Japan b School of Electronics and Computer Science, University of Southampton, SO17 1BJ, United Kingdom Received 28 February 2005; receivedin revised form 8 July 2006; accepted 15 July 2006

Abstract At present there exists a large gap in size, performance, adaptability and robustness between natural and artificial information processors for performing coherent perception-action tasks under real-time constraints. Even the simplest organisms have an enviable capability of coping with an unknown dynamic environment. Robots, incontrast, are still clumsy if confronted with such complexity. This paper presents a bio-hybrid architecture developed for exploring an alternate approach to the control of autonomous robots. Circuits prepared from amoeboid plasmodia of the slime mold Physarum polycephalum are interfaced with an omnidirectional hexapod robot. Sensory signals from the macro-physical environment of the robot are transducedto cellular scale and processed using the unique micro-physical features of intracellular information processing. Conversely, the response form the cellular computation is amplified to yield a macroscopic output action in the environment mediated through the robot’s actuators. © 2006 Elsevier Ireland Ltd. All rights reserved.
Keywords: Autonomous robots; Molecular computing; Coupled oscillators;Biologically inspired robotics

1. Introduction The prevalent approach to robot control is based on ‘behaviour decomposition’. The interaction loop between robot and environment is decomposed and treated as individual modules. This concept simplifies controller design and proved successful in many robotic systems including humanoid robots (e.g., Fujita et al., 2003). With ‘behaviourdecomposition’, the flexible selection of modules is critical for an adaptive and autonomous behaviour of the robot. One approach to address the decomposition problem disaggregates learning and interaction with the environment. While in learning mode, the robot learns or self-organises a proper module in terms of some environmental signal provided by a teacher, and subsequently uses this module in interactionmode.

∗

Corresponding author. E-mail address: 026d874n@stu.kobe-u.ac.jp (S. Tsuda).

The use of ‘behaviour decomposition’ enables robots to successfully work in either a stable work space or with the support of a teacher. The scheme can be implemented in localised (Tani and Nolfi, 1999) or distributed fashion (Tani et al., 2004). Without a teacher, however, delineating a boundary for theenvironment becomes an insurmountable challenge as the environment has no natural limit. In an unknown dynamic environment a disaggregation of learning mode and interaction mode is not useful unless this manifestation of the ‘frame problem’ (McCarthy and Hayes, 1969) can be overcome. In an attempt to address this difficulty, Brooks (1990) proposed the ‘subsumption architecture’ and implemented it ininsect-like robots (Brooks, 1991). This route, however, proved difficult to scale up because of computational cost. Quite a number of extensions and modifications of the above architectures have been reported, but the quest for genuinely autonomous robot control, capable of coping with an unknown complex environment, has not yet been successful. Nonetheless,

0303-2647/$ – see front matter © 2006Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.biosystems.2006.09.016

216

S. Tsuda et al. / BioSystems 87 (2007) 215–223

organisms seemingly do not face the difficulties encountered with robots. Cariani (1992) points to the plasticity of architecture and its ability to form appropriate sensory, computational, and effector structures in response to direct interaction with the...