Creating Intelligent Robots
Robots are becoming increasingly common, from the factory robots that build cars to the vacuuming robots that clean houses. For many applications, robots don’t have to be very smart. However, for others, they must operate autonomously and be capable of intelligent, flexible behavior. Examples of this include ocean exploration, long-term data gathering in changing environments, space exploration, and some military applications.
Robots are one kind of agent, that is, an entity that senses its environment (e.g., via sensors), makes a decision about how to behave, and takes actions (e.g., via “effectors”). Artificial intelligence (AI) research on giving agents intelligence, then, will lead to smarter, more capable robots.
The Orca Project
The Orca Project focuses on intelligent agent control. The application domain is autonomous underwater vehicles (AUVs) and other real-world agents. Orca is a schema-based reasoner, which means that it brings to bear packets of procedural and contextual knowledge – schemas – to determine how best to behave for a given mission and situation. It is an adaptive reasoner, which means that it is able to plan, yet modify its plans when the situation changes.
Current research on Orca focuses on two major topics. First, how can an agent commit appropriately to future behavior, while remaining ready to add new goals and revise its plans when the situation changes? Second, how can a reasoner apply what it knows about problem-solving contexts to allow it to automatically behave appropriately for new contexts?
CODA is a project of the Maine Software Agents/Artificial Intelligence Laboratory (MaineSAIL).