Evolution of functional specialization in a morphologically homogeneous robot

Abstract

A central tenet of embodied artificial intelligence is that intelligent behavior arises out of the coupled dynamics between an agent's body, brain and environment. It follows that the complexity of an agents's controller and morphology must match the complexity of a given task. However, more complex task environments require the agent to exhibit different behaviors, which raises the question as to how to distribute responsibility for these behaviors across the agents's controller and morphology. In this work a robot is trained to locomote and manipulate an object, but the assumption of functional specialization is relaxed: the robot has a segmented body plan in which the front segment may participate in locomotion and object manipulation, or it may specialize to only participate in object manipulation. In this way, selection pressure dictates the presence and degree of functional specialization rather than such specialization being enforced a priori. It is shown that for the given task, evolution tends to produce functionally specialized controllers, even though successful generalized controllers can also be evolved. Moreover, the robot's initial conditions and training order have little effect on the frequency of finding specialized controllers, while the inclusion of additional proprioceptive feedback increases this frequency.

The Utility of Evolving Simulated Robot Morphology Increases with Task Complexity for Object Manipulation

“…It has been demonstrated that indirect methods may produce phenotypes that grow in complexity without requiring a corresponding increase in genotypic complexity [9], and that control can be simplified if an appropriate morphology is found [29,24]. More recently it has been shown that a careful choice of morphology allows parts of the body to functionally specialize, rather than its being assumed a priori which parts of the body will perform which functions [2].…”

Evolution of an amphibious robot with passive joints

“…A variety of approaches have yielded controllers adapted to tasks, including evolution of neurocontrollers for walking [9]- [17], grasping an object [18], control of finless rockets [19], swimming [1], [8], station keeping [20], and complex behaviors such as foraging and game playing [7], [21]- [29]. While many studies focus on evolving controllers for fixed morphologies [2], [3], [5]- [7], [30]- [34], several recent investigations have addressed the evolution of morphology [8], [35]- [38], and even the coevolution of morphology and control [13], [15], [33], [39]- [42].…”

Evolving Monolithic Robot Controllers through Incremental Shaping

“…Another recent publication [2] used a similar experimental framework as the work just discussed to investigate a different problem. In this case the research question was not about the order in which the behaviors should be learned, but about what variables influence the frequency of finding functionally specialized controllers -that is, controllers that devoted part of the robot's body (it's front legs) to a single behavior (object manipulation) rather than using that body part for multiple behaviors.…”

“…Some have argued [8,10] that controllers should be organized in a modular fashion such that different control components are responsible for different behaviors, but is this modularity necessary? Recent work by our group and others has demonstrated that in fact, no, structural modularity is not always necessary [2,3,7,14]. An example of how a monolithic (non-modular) controller can be evolved to enable a virtual autonomous robot to perform a non-trivial sequence of behaviors will be presented in the next section.…”

“…But, what if this assumption is relaxed? In another recent publication [2], we demonstrated how a robot could be trained to locomote to and manipulate an object while the assumption of specialization of different body parts is relaxed: the robot had a segmented body plan in which the front segment was able to participate in locomotion and object manipulation, or it might have specialized such that it only participated in object manipulation. In this way, selection pressure dictated the presence and degree of specialization of the robot's morphology rather than enforcing such specialization a priori.…”