BECCA: Reintegrating AI for Natural World Interaction

摘要

Natural world interaction, the pursuit of arbitrary goals in unstructured physical environments, is an excellent motivating problem for the reintegration of artificial intelligence. It is the problem set that humans struggle to solve. At a minimum it entails perception, learning, planning, and control, and can also involve language and social behavior. Although it may be impossible for one agent to perform well in the entire problem space of natural world interaction, an agent’s fitness is indicated by being able to perform a wide variety of tasks. In order to address the problem of natural world interaction, a brain-emulating cognition and control architecture (BECCA) was developed. It uses a combination of feature creation and model-based reinforcement learning to capture structure in the environment in order to maximize reward. BECCA avoids making common assumptions about its world, such as stationarity, determinism, and the Markov assumption. BECCA has been demonstrated performing a set of tasks which is nontrivially broad, including a vision-based robotics task. Current development activity is focused on applying BECCA to the problem of general Search and Retrieve, a representative natural world interaction task.