Enabling Advanced Robots to Intelligently Operate

Tiffany Cappellari, Research Engineer, Southwest Research Institute

Modern robotics solutions seem to be becoming increasingly "intelligent" in their ability to perceive their environment and act accordingly. The ultimate goal of a robotic system is one that programs itself without any human intervention required. One where you can put a never before-seen part in its vicinity, tell it to start, and then come back an hour later with a completed inspection, sanding, painting, or any other desired process. Many of the latest open-source software packages are starting to enable some of these capabilities. This talk will go over a step-by-step process of how a system can go from having no knowledge of a presented part to safely executing a precise, desired process with little to no human intervention. Also, it will cover what application spaces are appropriate for such a technology, given the current state of robotics. The first step of the process is observation. The observation of an unknown part can be done using techniques of 3D mesh reconstruction from raw 3D sensor data. This mesh can then be filtered with intelligently designed processes. Next, using pre-configured toolpath parameters, evenly spaced, application-specific toolpaths can be generated on the mesh to accomplish the desired goal. Also, various machine learning vision techniques can be used to perform a feature extraction that can be tailored specifically for a given process to create desired toolpaths. Once toolpaths are generated a novel, safe, and efficient robot trajectory can be generated to execute the process without any offline programming required. Finally, the robot executes the process while continuously monitoring the environment. Continuous feedback on the state of the part could be used to inform decisions on future motions of the robot, minimizing the human intervention required.

Back