Q-Free Walk Ant Hybrid Architecture for Mobile Robot Path Planning in Dynamic Environment

摘要

This work provides a novel hybrid system for the trajectory planning of an autonomous mobile robot named Q-Free Walk Ant Hybrid Architecture (Q-FWAHA). The proposed approach combines two layers: a deliberative layer (global planning) and a reactive one (local planning), it is adapted to partially unknown environments (dynamic). It includes both reactive planning methods based on the Perception & Action principle (Sense & Act) and deliberative methods based on the Perception-Planning & Action principle (Sense-Plan & Act). The deliberative layer consists of a global planning algorithm and a model (a map) based on current knowledge of the environment. It is based on a method inspired by nature called Ant Colony Optimization (ACO) for optimal trajectory planning in known environments. The next layer is the reactive layer, it allows the robot to avoid collisions with dynamic obstacles by directly using sensory data. The reactive planning method used is a reinforcement learning approach for planning in dynamic environments; The Q-learning method. The deliberative layer is compared alone with an existing system for path planning, results show that the final path produced by our system is more efficient in terms of safety and energy saving. Simulations and experimental results to validate the proposed hybrid system are presented.