A Neurocognitive Approach to Robotic Cause-Effect Reasoning During Learning
Abstract
The growing importance of creating robotic systems for deployment in naval operations makes the development of more effective and efficient procedures for training them on new tasks a critical issue. A major problem that is faced in doing this is that current robotic systems are very limited in their ability to function autonomously, reason about unexpected events, and to learn rapidly to perform tasks. In this context, the primary goal of our proposed research is to develop a purely neurocomputational control system for humanoid robots that supports high level cognitive activities such as procedure learning, logical reasoning, goal-directed behaviors, and planning. Achieving this goal has substantial naval relevance: It is directly relevant to the development of innovative humanoid robots that can serve as an alternative shipboard/shipyard workforce for inspection, assembly, and maintenance tasks.There are three objectives in this work. 1. Create a neural virtual machine (NVM) that provides a purely neurocomputational framework for implementing cognitive-level algorithms that are currently readily implemented via more traditional symbolic AI methods, but much less so via existing neural network methods. 2. Use the NVM to implement a goal-directed, purely neurocomputational system for robotic imitation learning based on cause-effect knowledge, experimentally evaluating the resulting system using multiple inspection and maintenanceapplication tasks. 3. Conduct complementary experimental studies with human participants who learn the same tasks as the neurocomputational system, to provide new ideas to incorporate into the learning algorithms of robotic control systems.Our technical approach begins with the development of a general purpose software tool, the NVM, for rapidly embedding procedural knowledge into purely neurocomputational robotic control systems. Subsequently, we will use the NVM to implement a neurocognitive robotic controller for imitation learning modeled after CERIL, an existing imitation learning system that we developed recently with ONR support. Concurrently, we will conduct experiments with human subjects learning to perform the same tasks as our robotic system to gain insights that can be incorporated in our robotic system. The primary anticipated outcome and impact of our proposed research will be the development of not only a specific neurocognitive robotic control system for imitation learning, but also of a general purpose tool (the NVM) that can be used to rapidly develop future control mechanisms in a broad range of autonomous systems. Ourproposed work is the first effort to create an AI system for robotic imitation learning based on cause-effect reasoning that is completely directed by neurocomputational mechanisms, and it avoids the massive labeled data requirements and computational costs associated with current deep learning systems. Our research team at the Maryland and Syracuse will collaborate withengineers at Virginia Tech who are currently developing demonstration shipboard robotic systems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 23, 2019
- Source ID
- N000141912044
Entities
People
- James Reggia
Organizations
- Office of Naval Research
- United States Navy
- University of Maryland