Date of Award
Master of Engineering Science
Electrical and Computer Engineering
Dr. Mehrdad R. Kermani
Moving a robot between two configurations without making a collision is of high importance in planning problems. Sampling-based planners have gained popularity due to their acceptable performance in practical situations. This body of work introduces the notion of a risk function that is provided using the Support Vector Machine (SVM) algorithm to find safe configurations in a sampled configuration space. A configuration is called safe if it is placed at maximum distance from surrounding obstacle samples. Compared to previous solutions, this function is less sensitive to a selected sampling method and resolution. The proposed function is first used as a repulsive potential field in a local SVM-based planner. Afterwards, a global planner using the notion of the risk function is suggested to address some of the shortcomings of the suggested local planner. The proposed global planner is able to solve a problem with fewer number of milestones and less number of referrals to the collision detection module in comparison to the classical Probabilistic Roadmap Planner (PRM). The two proposed methods are evaluated in both simulated and experimental environments and the results are reported.
Anooshahpour, Mahdi, "A Motion Planner For Robot Manipulators Based on Support Vector Machines" (2011). Digitized Theses. 3242.