Comparison of behavior-based and planning techniques on the small robot maze exploration problem

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Miniature mobile robotic agents often have limited sensory capabilities, posing challenges for effective control.
• Behavior-based and planning control architectures represent distinct paradigms for robot navigation and task execution.

Purpose of the Study:

• To compare the efficacy of behavior-based control mechanisms against a localization and planning approach for miniature mobile robots.
• To evaluate two specific behavior-based methods: a radial basis function neural network trained via evolutionary algorithms and a traditional reinforcement learning algorithm.

Main Methods:

• Implemented and compared a behavior-based control system utilizing a radial basis function neural network trained with an evolutionary algorithm.
• Implemented and compared a behavior-based control system using a traditional reinforcement learning algorithm over a finite state space.
• Developed and evaluated a robot control architecture based on localization and planning for comparative analysis.

Main Results:

• Behavior-based control, specifically using evolutionary algorithms for neural network training, demonstrated comparable performance to planning-based methods.
• Reinforcement learning within a finite state space also provided a viable behavior-based control alternative.
• The study highlights the potential of reactive, behavior-based strategies in resource-constrained robotic systems.

Conclusions:

• Behavior-based control methods, including those employing neural networks and reinforcement learning, are effective alternatives to planning-based approaches for miniature mobile robots with limited sensing.
• The choice between behavior-based and planning approaches may depend on specific task requirements and available computational resources.
• Further research can explore hybrid approaches combining the strengths of both paradigms.