Performing a Cleaning Task in a Simulated Human-Robot Interaction Environment

In this work we introduce a human-robot interaction task where a robot performs a domestic activity which consists of cleaning a table in a simulated environment. Our scenario has two objects (sponge, cup) and the robot is allowed to perform four actions: get <object>, drop <object>, go <location>, and clean. For this purpose, we develop an interactive reinforcement learning approach integrated with the DOCKS speech recognition system to provide human interactive feedback via speech guidance. Our approach is written in Python and connected to the V-REP simulator using the remote API. The external trainer is able to advise the robot on what action to perform next using live or previously recorded instructions. Once having received the external advice, a Baxter robot performs the next suggested action if possible. To be able to move to a location or clean the table, direct planning is employed; for getting or dropping an object, the robot uses inverse kinematics which are integrated in the simulator. The V-REP simulator currently lacks simulated sound sources which could be an interesting extension for further experiments. Nevertheless, by using our human speech guidance architecture we are able to finish the task faster than using autonomous learning.

@InProceedings{CTW15, 
 	 author =  {Cruz, Francisco and Twiefel, Johannes and Wermter, Stefan},  
 	 title = {Performing a Cleaning Task in a Simulated Human-Robot Interaction Environment}, 
 	 booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems, Workshop An Open-source Recipe for Teaching/Learning Robotics with a Simulator},
 	 editors = {},
 	 number = {},
 	 volume = {},
 	 pages = {},
 	 year = {2015},
 	 month = {Oct},
 	 publisher = {},
 	 doi = {}, 
 }