Future Deliverables 1999 - 2000 | 2000 - 2001 2001 - 2002: Learned Coordinated Body Movement Human-like dynamics require joint control that is responsive to physics of the world and robot body structure. This aspect of control will be acquired through Cog exploring and learning about its workspace. Cog will learn refined coordination of its torso, arm and hand movements using a layered approach of reflex modeling, reflex inhibition and multi-joint simulated muscles. Object Preferences and Recall Cog will habituate, obey commands to attend, ignore or engage objects using task-driven saliency. Cog will remember objects indexed by location and visual appearance. It will be commanded to return its attention to previously observed objects and locations. Mirror Neuron Model Acquisition Cog will perform an action in the world while simultaneously its mirror neuron model will incorporate the sensorimotor aspects of the action (visual data and motor command sequences) into a representation of the causality between the action and its consequences. Later, when observing another person performing some action, Cog will enact a motor sequence associated with the match of that action with one in its repertoire that is likely to have the same consequence. Precise Control in the Presence of Noise When controlling a robotic arm large state errors are introduced by gravity and nonlinearities. Despite the presence of noise, Cog's arm will precisely operate under parametric and non-parametric adaptive control using a nonlinear sliding-modes controller. Robust, Continuous Robot Operation A robot will continuously sense and act in its environment without being shut down. It will regulate its use and acquisition of energy and operate according to the drives stipulated in a motivational system.