Quantifying Morphological Computation in a Legged Robot using KL Devergence #

Vijay Chandiramani

10:50 Tuesday in 4Q04.

Part of the Complex systems and control session.

Abstract #

Legged robots are required to be compliant to handle perturbations, such as impulsive contact forces from differ- ent terrains, or from other obstacles. In that respect, mod- elling system dynamics using active control becomes chal- lenging due to uncertainty and with parameters changing over time due to operational and environmental conditions. Biological systems use an inter-play of active control and using the body as a computing resource to adapt to such changing environment. Designing legged robots using such an inter-play remains under-explored [1]. This paper, as part of wider research on designing more adaptive robots using this inter-play, demonstrates the quantification of the extent of such embodiment using Bayesian theory and Kulback divergence method, which can be used as a useful input in the design process.