Automatic object searching by a mobile robot with single RGB-D camera

Automatic object searching is one of the essential skills for mobile robots to operate in unstructured and dynamical environments. It requires a robot to be endowed with object identification, obstacle avoidance, path planning, and navigation abilities. In this paper, a generic framework for automatic object searching is proposed. It has obstacle avoidance capability in unstructured and dynamical environments assisted by a behaviour learning algorithm using deep belief networks (DBN). As soon as a target object is recognized by a vision-based method, a bug-based path planning algorithm will be triggered for the robot in order to approach the target. Compared to state-of-art systems where laser sensors and cameras are widely used together for object identification, obstacle avoidance, path planning and navigation, only a single low cost RGB-D camera is used in our system to perform the above tasks. The proposed system has been implemented and tested in two indoor environments including a laboratory and a pantry of an office. The experimental results demonstrate the effectiveness of the proposed framework.