Tutorial on Embodiment

3.2.3.2. Exploiting self-adaptation: cockroaches climbing over obstacles*

 

Cockroaches cannot only walk fast over relatively uneven terrain, but they can also, with great skill, negotiate obstacles that exceed their body height. They have complex bodies with wings and three thoracic segments, each one with a pair of legs. There is one thoracic ganglion at each segment (i.e., prothoracic, mesothoracic, and metathoracic, see Fig. 3.2.3.2.1) controlling each pair of legs. Although both the brain and the thoracic ganglion have a large number of neurons, only around 250 neurons - a very small number - descend from the brain to the body (Staudacher 1998). Each leg has more than seven degrees of freedom (legs and feet are deformable to various extents), some joints in the legs are very complicated and there is also an intricate set of joints in the feet.

 

Fig. 3.2.3.2.1. Schematic representation of cockroach anatomy. As the cockroach is climbing over an obstacle, the configuration of the middle (mesothoracic) shoulder joint is reconfigured. The region where the change is taking place is roughly marked with a red rectangle (picture courtesy Roy Ritzmann, Case Western Reserve University).


Now, how is it possible to control all these degrees of freedom with such a small number of descending neurons? In what follows, we discuss one potential solution that, although entirely hypothetical, does have a certain plausibility and it further illustrates the idea of morphological computation.

Suppose that the animal is walking on the ground using its local neural circuits which account for stable forward locomotion. If it now encounters an obstacle - whose presence and approximate height it can sense by means of its antennae. The brain, rather than changing the movement pattern by evoking a different neural circuit, "re-configures" the shoulder joint (i.e., thoracic-coxa, see Fig. 3.2.3.2.2) by altering local control circuits in the thoracic ganglion. As a result, the middle legs are rotated downward so that extension now generates a rearing motion that allows it to easily place its front legs on top of the block with typical walking movements (Fig. 3.2.3.2.2 and Video 3.2.3.2.1). The local neural circuits continue doing essentially the same thing - perhaps with increased torques on the joints - but because now the configuration of the shoulder joint is different, the effect of the neural circuits on the behavior, the way in which the joints are actuated, changes (See Watson & Ritzmann 1998; Watson et al. 2002).

Fig. 3.2.3.2.2. Cockroach climbing over obstacle. (a) touching block with antenna. (b) front leg put on block, middle legs pushing down, posture changing to reared-up position. (c) front legs pushing down on block, middle leg fully stretched (picture courtesy Roy Ritzmann, Case Western Reserve University).

 

Video 3.2.3.2.1. Cockroach climbing over obstacle. (courtesy Roy Ritzmann)

 

Interestingly, this change is not so much the result of a different neural circuit, but of an appropriately altered joint morphology. Instead of manipulating the movements in detail, the brain orchestrates the dynamics of locomotion by "outsourcing" some of the functionality to local neural circuits and morphology through an ingenious kind of cooperation with body and decentralized neural controllers. In order to modify the morphology, only a few global parameters need to be altered to achieve the desired movement. This kind of control through morphological computation has several advantages. First, the control problem in climbing over obstacles can be solved efficiently with relatively few neurons. Second, this is a "cheap solution" because much of what the brain would have to do, is delegated to the morphology, thereby freeing it from unnecessary control tasks. Third it takes advantage of the inherent stability of the local feedback circuits rather than working in opposition to them. And fourth, it illustrates a new way of designing controllers by exploiting mechanical change and feedback (See the principle of "cheap design", Pfeifer et al., 2007).

 

* This case study has previously appeared in Pfeifer & Gomez, 2009.

References:

Pfeifer, R., Lungarella, M., Iida, F.: Self-organization, embodiment, and biologically inspired robotics. Science 318, 1088-1093 (2007)
Pfeifer, R. & Gomez, G. (2009). Morphological computation - connecting brain, body, and environment. In B. Sendhoff, O. Sporns, E. Körner, H. Ritter, & K. Doya, K. (eds.), Creating Brain-like Intelligence: From Basic Principles to Complex Intelligent Systems (pp.66-83). Berlin: Springer.
Staudacher, E.: Distribution and morphology of descending brain neurons in the cricket gryllus bimaculatus. Cell Tisues Res. 294, 187-202 (1998)
Watson, J., Ritzmann, R.: Leg kinematics and muscle activity during treadmill running in the cockroach, blaberus discoidalis: I. slow running. J. Comp. Physiol. A 182, 11-22 (1998)
Watson, J., Ritzmann, R., Pollack, A.: Control of climbing behavior in the cockroach, blaberus discoidalis. ii. motor activities associated with joint movement. J. Comp. Physiol. A 188, 55-69 (2002)

 

<<PREVIOUSHOMENEXT>>