Autonomous navigation in GPS denied environments using MPC and LQR with potential field based obstacle avoidance

Abstract

Workers in mining industry are posed with hazardous environments due to the nature of the work inside a mine. Gas leaks, explosions, rock falls, entrapment, long exposure to dust are all potentially fatal conditions for the workers. Although solutions for the problems are being implemented, they are not sufficient and mostly are very expensive. \newline Autonomous robots can reduce the risk for miners by taking over potentially dangerous tasks for them. For instance, an autonomous robot can carry operations like air quality assessment, inspection of dangerous mine conditions and even perform search and rescue tasks in disaster situations. \newline This thesis presents robots that can traverse through the mine environment with on board sensors collecting data without any human intervention. Control and Obstacle avoidance algorithms are designed and presented in this thesis for the robots, ground and aerial platforms. A Model Predictive Control (MPC) approach is presented which includes pre-packing of necessary terms that would help decrease the computation costs. A Linear Quadratic Regulator (LQR) approach is also presented and its performance against the Model Predictive Control approach is presented in presence and absence of obstacles. A Potential Fields based obstacle avoidance approach is presented which makes use of octomaps. \newline Experimental results are promising as both aerial and ground platforms perform navigation without any GPS and avoid obstacles if any, in a simulation. Fast solve times on the order of hundreds of micro seconds are obtained and the results are compared with other existing techniques and presented. A real-time implementation of the ground robot has been made in various GPS denied environments and the results are presented. In real-time as well, the robot performs navigation avoiding obstacles in all cases.