Anizotropik olarak yerleştirilmiş yumuşak omurgalar sayesinde manevra kabiliyeti yüksek çok ayaklı hareket yapabilen tek aktuatörlü modüler minyatür robot

This thesis presents the design, modeling, and control of a modular miniature robot with a single actuator. The robot has functionally diverse passive legged modules that are connected by anisotropically positioned soft I-beam PDMS backbones. These provide directional compliance and combined with controlled vibration by an unbalanced rotating mass, they allow generating multiple gaits in the legs that cause forward, lateral, and rotational locomotion without the use of additional actuators or added mechanics. A scalar parameter, $\alpha$, was incorporated to define the orientations of the backbones. It enabled efficient characterization of locomotion at various frequencies of actuation. The robot has a high-fidelity dynamic model in which the flexibility in the backbones was simulated by multi-segmented revolute joint chains. The joint parameters are obtained by performing a series of oscillation experiments in the physical system and optimized using Bayesian optimization. An effectiveness index was assembled based on the system robustness alongside sensitivity to variation in friction. Additionally, the control signals to be applied by the rotation of the system around the robot geometry center, in order to generate 2D locomotion, are learned using a reinforcement learning algorithm. The outcome of the behavior was verified in simulation, and applied to maneuver through a tight path. Overall, this work demonstrates that anisotropic and asymmetric mechanical design, in conjunction with learning-based control, enables adaptive locomotion in underactuated systems using minimal hardware. The outcome instigates the development of compliant robots that can scale for deployment in constrained, alongside unstructured, environments.