Current work focuses on determining methods of integrating multiple diverse sensors, and corresponding motor states, into a unified framework for learning and control. Initial implementation will be performed on the EvBot II robotic platform. Through effective sensorimotor integration, learning time will be reduced, and the learned control systems will exhibit more robustness of control and increased generalizability. Also, by combining and correlating multiple sensors and motor states, the system will be increasingly immune to noise, and corrupted or missing data from the integrated sensors.
In addition, work is being done to expand the sensorimotor integration to include sensors residing on other robots in the colony, and from surrounding independent sensors that make up a distributed sensor network. Using sensors on other robots in the colony and the distributed sensor network enables the EvBots to maintain a common expanded knowledge of the environment in which they are interacting. This is essential for effectively coordinating efforts between distributed, possibly diverse, groups of autonomous robots.
In order to facilitate this full and complete sensor integration, a modular control environment is being developed. This modular control environment will increase the speed and ease with which new sensorimotor elements can be prototyped and evaluated on the mobile robotic platform. Due to its modular nature, this control environment will be easily portable to different robotic, or simulation, platforms.
The combination of these resulting technologies will enable the autonomous robotic system to adapt to constantly changing environments and sensorimotor systems. Such a system will exhibit the necessary robustness of control and behavior that is critical in real world situations and environments.