The development of an aerial robotics laboratory highlighting the first experimental validation of optimal reciprocal collision avoidance.

摘要

This thesis details the development of the Algorithmic Robotics Laboratory, its experimental software environment, and a case study featuring a novel hardware validation of optimal reciprocal collision avoidance. We constructed a robotics laboratory in both software and hardware in which to perform our experiments. This lab features a netted flying volume with motion capture and two custom quadrotors. Also, two experimental software architectures are developed for actuating both ground and aerial robots within a Linux Robot Operating System environment. The first of the frameworks is based upon a single finite state machine program which managed each aspect of the experiment. Concerns about the complexity and reconfigurability of the finite state machine prompted the development of a second framework. This final framework is a multimodal structure featuring programs which focus on these specific functions: State Estimation, Robot Drivers, Experimental Controllers, Inputs, Human Robot Interaction, and a program tailored to the specifics of the algorithm tested in the experiment. These modular frameworks were used to fulfill the mission of the Algorithmic Robotics Lab, in that they were developed to validate robotics algorithms in experiments that were previously only shown in simulation.;A case study into collision avoidance was used to mark the foundation of the laboratory through the proving of an optimal reciprocal collision avoidance algorithm for the first time in hardware. In the case study, two human-controlled quadrotors were maliciously flown in colliding trajectories. Optimal reciprocal collision avoidance was demonstrated for the first time on completely independent agents with local sensing. The algorithm was shown to be robust to violations of its inherent assumptions about the dynamics of agents and the ability for those agents to sense imminent collisions. These experiments, in addition to the mathematical foundation of exponential convergence, submits that optimal reciprocal collision avoidance is a viable method for holonomic robots in both 2-D and 3-D with noisy sensing. A basis for the idea of reciprocal dance, a motion often seen in human collision avoidance, is also suggested in demonstration to be a product of uncertainty about the state of incoming agents. In the more than one hundred tests conducted in multiple environments, no midair collisions were ever produced.