Robotic swarm inspection for structures offers a flexible, scalable, and cost-effective solution in comparison to human inspection such as the benefit of being resilient to individual failures, larger general coverage, and flexibility in sensing. One novel destination for such applications is space. These robot swarms could be deployed in on-orbit satellite hulls, human space capsules, or future inhabited space structures. With routine monitoring in the harsh space environment, this solution offers a safe alternative against the risk in human inspections. This work moves towards this reality in three major points. First, this work presents a physics based simulation of a Bayesian inspection algorithm for randomized grid surfaces through a swarm of small-scale wheeled robots. These robots have the capability of moving on unique 3D surfaces, communicating with each other, and sensing vibrations. We deploy these robots to inspect a two-outcome surface of black and white grids in simulation, where we find that the algorithm performs well. Second, we introduce an optimization framework based off particle swarm optimization that is deployable on Amazon Web Services' multi-node clusters. We use this framework to find a set of heuristic parameters that show improvements in our defined criteria. The optimization framework reveals tradeoffs between accuracy and speed, while maintaining a minimum needed surface coverage. Third, we introduce novel hardware development that enables collision avoidance for physical deployment and experiments. With this, the work sets up the pathway to bring optimized algorithm parameters into a close relationship with hardware experiments.
Darren Chiu, UG '23:
https://www.linkedin.com/in/darren-chiu-ee/ +
https://darren-chiu.github.io/