Tuesday, 3 February 2015

Are robots embodied?

‘Are robots embodied’ is the question posed by Hooijmans & Keijzer in Robotics, Biological Grounding and the Fregean tradition. They describe the ‘symbol grounding’ problem where symbols are not related to the actual world. They need sensors and actuators that embodiment and situatedness provide to become grounded. A similar ‘biological grounding’ question exists for robots where it cannot be sure that robots are really agents in the same way as organisms. 

They quote Brook's definition of embodiment as ‘intelligence derived from a specific bodily system involving sensor and actuators’ and of situatedness as ‘intelligence arises through interaction of an embodied agent with a particular environment.’  Dynamical systems theory is added to this model to give DESC – dynamic, embodied and situated cognition.

Pfeifer argues that the best way to build robots is from an embodied point of view in On the role of embodiment in the emergence of cognition. Robots in the real world, especially those that interact with us, will need to be soft – soft to touch, soft skin, soft movement and to be able to deal with emotions. An understanding of embodiment will be required to achieve this.

He shows how the passive dynamic walker can walk down a slope without any motor or control system in a most natural way. It demonstrates how a system can emerge without central control and purely from the interaction of the body and environment. The memory and processes are distributed between the body and environment. He argues that the design of robots should start from the body. 

He is not arguing that the brain should be ignored. Learning needs to take place in an embodied fashion. There is a task distribution between the brain, body and environment. Movement is not controlled but orchestrated. The movement of the legs has been outsourced to the body as in the passive dynamic walker.

The approach of engineers such as Pfeifer will ensure that the embodiment of robots will be achieved according to the definition of DESC. However the biological grounding question remains. Luckily Dilbert has already made a start on this.

No comments:

Post a Comment