Evaluating Cognitive Architectures for Intelligent and Autonomous Unmanned Vehicles

Unmanned vehicles (air-, ground-, and sea-based), or mobile robots, have become increasingly important for both civilian and military applications and both intelligence and autonomy are important for these vehicles and their continued success. Many of these vehicles, however, currently require significant human supervision. For example, the Predator and GlobalHawk are essentially remotely piloted aircraft and the ground vehicles that completed the DARPA Grand Challenge displayed impressive autonomy, but had very little onboard intelligence. Under-water vehicles and mobile robots used for planetary explorations could also benefit from increased intelligence and autonomy due to the delays and limited bandwidth typical of communication with these robots. Using cognitive architectures in these systems presents an intriguing possibility for developing unmanned vehicles that can behave both intelligently and autonomously. It is extremely difficult, however, to compare the use of cognitive architectures for unmanned vehicles since the architectures have seldom been applied to the same problem or vehicle. Most of them are very difficult to use and different architectures may not be well suited to work on the same problems. This represents a significant barrier for someone who has limited experience with cognitive architectures and is interested in selecting an architecture to implement on an unmanned vehicle. We recently reviewed several possible software systems (Long et al. 2007) for unmanned vehicles. We have also recently documented our use of an intelligent controller in unmanned air vehicles (UAVs) (Sinsley et al. 2007, Miller et al. 2007).