Developing a Robotic Simulator and Video Games for Professional and Public Training
USDOT #69A3551747126
Abstract
This project developed a simulation training software for operators to monitor and supervise autonomous robots when inspecting steel truss bridges. Since the software abstracts the command and situation visualization interface when dealing with simulated inspection robots and bridge, the same software can be connected to physical robots to control and monitor physical robots during bridge inspection. Version 1 of Simulation Training and Control (STACS) was developed and tested with two virtual bridges and one virtual climbing robot for steel truss member inspection and one virtual flying UAV for visual inspection. This version of STACS also prototyped and tested a Robot Operating System (ROS) bridge to a physical Roomba robot to demonstrate teleoperative control of the physical robot and the ability to send commands and receive sensor information as well as to stream camera views. Initial experiments with STACS in simulation revealed the importance of robot routing to minimize inspection time while visiting every member of the bridge truss. Although complete, optimal routes for one climbing robot can be found in polynomial time, k-robot complete optimal routing, corresponding to the MinMax k Chinese Postman Problem, is NP-complete and no reasonable time optimal algorithm exists. A genetic algorithm was developed to near-optimally solve the k-robot problem and was scaled up caching and by parallelizing the genetic algorithm to use GPUs. Tests on benchmarks and a test bridge validated and demonstrated the feasibility of our approach for k-robot routing and linear speedup as k increases.
