In the field of mobile robotics, there is a substantial challenge surrounding reliable navigation of unstructured environments with irregularities. A robot capable of versatile and robust locomotion is required, in order to handle randomly cluttered environments, which may also vary in substrate. This problem has two interdependent components; the physical form of the robot, and its mode of locomotion, or gait. By observing nature, researchers can find inspiration in organisms that have spent decades of evolution becoming expertly adept in particular tasks, and manifest this through behaviours and physical features.

As a species, snakes are fascinatingly versatile at locomoting; depending on the substrate and irregularities of the environment, various gaits are used. The physical features of biological snakes are also diverse; microscopic ventral scale structures vary with habitat and primary gait. These scale structures impact the frictional properties of the ventral scales, resulting in varying degrees of anisotropy. The physical form of biological snakes is also attractive from a design perspective; a slender form, instrinsic stability and hyper-redundancy make a snake-inspired robot (snake robot) ideal for navigating cluttered environments. A snake robot able to harness the diverse set of available snake gaits presents an appealing solution to versatile locomotion. The aforementioned relationships between substrate, gait and ventral scale structure are conflicting, however; for a snake robot to perform all snake gaits, it must also possess ventral scales that can vary in structure.

This thesis presents Noodle; a cost-effective snake robot with modular ventral panels on the sides of its body. This enables configuration of the ventral contact surfaces prior to deployment onto a given substrate, and the ability to ‘select’ its ventral side simply by rolling. By modularising the contact surfaces into these panels, access to varying degrees of frictional anisotropy is granted simply by swapping them. The cross-section of Noodle ressembles a chamferred square; thus, there are up to 4 unique sets of panels able to equipped at a time. Two panel designs are 3D printed from PLA, to assess the effectiveness of altering geometry to achieve frictionally isotropic and anisotropic designs. The anisotropic design is able to achieve 3 times more friction in the lateral direction than longitudinal when interacting with carpet, which served as a textured, slightly yielding surface.

The body of Noodle is modularised into identical single degree of freedom (DOF) segments, which may be appended to one another with the rotational axes in parallel, or alternating in pitch and yaw; these achieve the planar and 3D configurations, respectively. Through the inclusion of series elastic actuators (SEA), compliance and torque sensing are achieved. The SEA is realised by attaching a compliant element in between the servo motor and driven segment. A conventional hobby servo was used. Two designs of compliant element, or series elastic element (SEE), are presented: coil spring and elastomer, where the latter is used in the final design. The material properties of elastomers depend on temperature, while the elastomer itself exhibits dampening and hysteretic deformation; the coil spring SEE offers higher temperature invariance and no hysteresis, but meeting the stiffness requirements of the SEE were challenging. Furthermore, the elastomer SEE is much simpler to manufacture, and its shortcomings have been shown to be relatively minor in the use case of a snake robot SEE. The elastomer SEE is experimentally characterised to have a torsional stiffness constant of 3.54 [N·m/rad]; with a 12-bit rotary encoder resolution, a 0.005 [N·m] torque-measurement resolution is achieved. Each segment of Noodle consists of a custom embedded system, SEA, and 3D printed body. Special attachments to ressemble the head and tail are also included in the proposed design. These attachments, the segment bodies, and parts of the SEA, are 3D printed from PLA; a cost-effective manufacturing decision that also allows for rapid prototyping.

To validate the design and assess gait-execution capabilities, Noodle is configured in the planar configuration, with a total of 8 joints. The average cost per segment module is 158 NZD. In the planar configuration, with the joints acting in the pitching direction, the rectilinear gait can be implemented; the robot propels itself by executing successive poses that ressemble a dorsally-propagating wave. The impacts of the wave parameters on average movement speed are investigated; keeping temporal frequency constant (which will self-evidently affect movement speed) and using a range of values for amplitude and phase shift, speeds between 2.5±0.1 and 5.2±0.1 [cm/s] are achieved.

A control architecture and system are proposed, based on the embedded architecture and torquemeasuring SEAs. Local stiffness control (LSC) is implemented, which allows closed-loop control of the joint stiffnesses, which in turn achieves adaptive behaviour in response to environmental stimuli. The effectiveness of the LSC is evaluated across several experiments wherein Noodle successfully overcomes an obstacle approximately 70% of its own height, using the impedance variation of the LSC. The results show decreased time taken to conquer the obstacle with increasing torque control gain, to an extent; beyond a certain gain value, the decrements in the times taken become marginal. The implications of the control and embedded architectures on the control system responsiveness are also evaluated.

Noodle, with its modular ventral panels, poses an appealing solution to the problem of conflicting snake gaits, ventral surface properties, and traversed substrate. Preliminary planar control with joint torque feedback is demonstrated, paving the way for the implementation of sophisticated 3D gaits. Ultimately, these elements make Noodle an excellent intermediate platform in the pursuit of mobile robots capable of navigating unstructured environments with irregularities.