Toward Consistent Robotics Simulation Through Validation

• Main Author: Taylor, James R.
• Language: EN
• Published: The George Washington University 2018
• Subjects: Robotics|Computer science
• Online Access: http://pqdtopen.proquest.com/#viewpdf?dispub=10929572

<p>Multi-rigid body dynamics simulation supports design, prototype, test, and evolutionary stages of robotic development. However, the reliability of simulating tasks critical to robotics, i.e. grasping and locomotion, remains low and represents a barrier to future robotics applications. A lack of simulation reliability is typically exhibited by a simulated robot performing a task while under similar conditions the real system fails to perform that same task. Simulations that are not reliable may mislead roboticists into costly, infeasible designs or give false safety assurances of systems where safety is paramount, e.g. autonomous vehicles. Roboticists therefore need assurances that simulation will either consistently predict real performance or indicate that it can not make sound predictions. General software engineering proposes verification and validation as practices for quality assurance. In simulation software, verification checks that implementation is consistent with theory, and validation checks that simulation is consistent with real behaviors. While verification is critical, validation is necessary to objectively demonstrate simulation consistency. This thesis studies validation of multi-body rigid body dynamics simulations with contact and friction in an effort to improve the state of the art in robotics simulation. This work identifies scenarios relevant to robotics tasks, examines simulators performing these tasks, and identifies ways that current simulators should be used and future simulators should be improved. Additionally, this work identifies temporal consistency as a critical requirement for robotics simulators, proposes a temporally consistent simulator architecture, and studies the performance difference between temporally consistent and existing robotics simulators. Finally, this work studies the motion of a minimally simple, yet still sophisticated, real world robot that we can model and examines a means to validate simulation of the robot.