Sensor Fusion-Based Supervised Learning Approach to Developing Collaborative Manipulation System with Variable Autonomy

Abstract

The objective is to create and program a demonstration of a manipulator arm that is capable of detecting objects, distinguishing between them, and relocating target objects to corresponding locations. The experiments are carried out to evaluate and improve the arm’s design, and to gather preliminary data from the sensors to better program the system’s behavior. The proposed manipulation system design makes the robot arm capable of performing a predetermined series of object relocation tasks with or without outside (human) commands from an unknown initial state. During the experiments, the performance of the robotic manipulation system is compared between two separate sensing conditions: (i) use of an ultrasonic sensor alone, and (ii) use of an ultrasonic sensor plus a light sensor. The human operator has a varying role in the manipulation. The operator may reset system inputs, put command, observe the operation, and collaborate with the system as a complementary performer or co-worker when the system needs human’s support. A survey is conducted to determine the potential human involvement and human factors associated with the manipulation system, which helps determine and vary the autonomy levels of the system. Then, an experiment is conducted to determine the autonomy levels of the system based on assessing the varying contribution of the human operator. A supervised learning approach is then proposed that may learn from previous events, predict required autonomy levels for future events, and thus may help maintain appropriate autonomy levels with task requirements. The results may help develop intelligent robotic manipulation systems in industries that may work independently or in collaboration with human workers with varying levels of autonomy.