|Members - Faculty, students, and collaborators
|Research - Details about our current research projects
|Theses - Publications and code releases for student theses
|Publications - Recent technical papers and presentations
|Software - Recent software releases
Research in the IDEA Lab focuses on creating and enabling intelligent autonomous agents that can successfully interact and learn while embedded in the real world. Success in such a dynamic multi-agent environment requires the following four key components.
- Interaction: An agent must be able to interact with the environment by taking actions and observing the results of its action. The environment should provide feedback, either implicitly or explicitly.
- Discovery: An agent must discover the salient patterns from the overwhelming amount of information available to it. Success in complex environments requires an agent to focus its attention on the essentials.
- Exploration: An agent must balance the need to exploit its current knowledge about the task at hand with a need to actively explore alternative solutions. Curiosity is a required component of any successful agent.
- Adaptation: An agent must be able to adapt to changes in the environment as well as to changes within itself (such as hardware degradation).
We address each of these issues using methods drawn from all aspects of artificial intelligence, machine learning, knowledge discovery, data mining, and robotics. The agents that we study are embodied in both software and hardware. General areas of research interest include the following.
- Autonomous pattern discovery: We study methods to identifying salient patterns such as abstractions or testable hypotheses in complex data sets.
- Knowledge representation: Real-world data is complex, large, and dynamic. We are creating knowledge representations for spatially and temporally varying relational data where agents will be able to reason about objects and relationships as well as how the evolution of the environment over time and space.
- Robust machine learning techniques: We combine many approaches to machine learning to create a more robust learning algorithm. This includes techniques drawn from reinforcement learning, supervised learning, and relational learning.
Our real-world application areas provide an opportunity to make a significant difference outside of academia. The requirements imposed by real applications stimulate the development of new approaches. Current application areas follow.
- Severe weather prediction: In collaboration with OU's School of Meteorology, we are developing new spatiotemporal relational knowledge discovery methods for use on mesoscale meteorological data. These techniques will identify causes of severe weather events, including tornados and thunderstorms.
- Robotics: In collaboration with robotics researchers, we are developing novel approaches to autonomously creating robust task-oriented knowledge, enabling robotic manipulators to capture the essential structure of a task. This will enable a robot to generalize its knowledge to similar tasks and greatly enhance the possible tasks for mobile manipulation platforms.
In addition, we have a long-term interest in applications that will assist a permanent human presence in space.