Tuning behavior-based robotics in mixed reality using unmanned aerial vehicles

Abstract

This thesis investigate a behavior-based robotics (BBR) approach for enabling autonomous unmanned aerial vehicle (UAV) navigation and obstacle avoidance in unknown environments without relying on global maps. The system leverages lightweight reactive control architectures, finite state machine (FSM) and subsumption architecture, to support real-time decision making based on local sensor feedback. A hardware-in-the-loop (HIL) framework is employed to safely evaluate controller performance under realistic conditions. The HIL setup integrates the Crazyflie UAV platform, lighthouse-based state estimation, simulated sensor inputs, and a modular software stack based on microservice architecture. To further enhance navigation efficiently, genetic algorithm (GA) optimization is applied to tune key controller parameters, including safe distances, tuning angles, and behavior transition timings. Experimental results demonstrate that both FSMbased and subsumption-based controllers enable robust mapless navigation and effective obstacle avoidance in complex indoor scenarios. The subsumption controller exhibits superior performance in cluttered environments, while the FSM controller performs better in open spaces. The findings highlight the feasibility of behavior-based UAV control in GPS-denied, mapless indoor settings, and demonstrate the value of modular HIL testing for rapid prototyping and validation of autonomous navigation strategies.