Robust performance of autonomous robots in unstructured environments.

The problem of navigating in and interacting with an unstructured environment presents challenges to traditional learning and control approaches. However, the nature of emergency response situations requires that autonomous robots’ performance be robust to unmodeled environments and unexpected challenges. One approach to providing this capability is presented here: S-Learning. S-Learning, an experience-based learning algorithm, is implemented in the control of a seven degree-offreedom robotic arm. S-Learning stores sequences of discretized (discrete in time), quantized (discrete in magnitude), and categorical (uninterpreted) sensor data and actuator commands. Handling the data in this way removes explicit models about the environment, robot kinematics, dynamics, and structure. Instead, a bootstrapped model is generated on the fly by observing sequences of sensory and command events. S-Learning is based on a neuro-psychological model of learning and movement control in humans and seeks to mimic the strategies used by the brain to solve this problem.