Where to begin?
This research story started as an undergraduate thesis project in 2014, where I worked to design a rigid grasping mechanism for a robot to climb throughout truss structures such as power transmission towers. The aim of the overall robotic platform was to perform inspection and maintenance tasks, which highly trained human workers currently complete in often dangerous conditions (for example, while the high voltage transmission lines are still live!). My interest in this undergraduate thesis project quickly grew into a PhD thesis topic; researching soft robotic manipulators for this application.
Soft robotics is a relatively new area of robotics research and has gained much attention due to the benefits of compliancy and adaptability in soft "mechanisms" which are difficult to achieve in conventional rigid robotics. An advantage of compliancy in robotics is safety; humans can safely interact with soft or compliant structures, without the risk of being harmed during their operation. Many soft robotic designs are biologically inspired; with inspiration coming from completely soft structures such as worms, caterpillars, octopuses and fish, amongst many others. To mimic the operation of these natural structures, the design of soft robotic manipulators tend towards the use of hyperelastic materials such as silicones and polyurethanes, actuated using pneumatics, or hydraulics. A disadvantage of using these materials in robotic manipulation is a lack of output force as compared to a completely rigid mechanism. Furthermore, due to a lack of discrete and distinguishable Degrees of Freedom (DoF) in soft robotic structures, sensing and predicting their motion is extremely challenging.
The focus of my research has been towards the use of a soft, compliant grasping mechanism capable of adapting to the varying cross-sectional shapes and sizes of structural beam members in power transmission towers. First, I designed a soft grasping platform as the basis of my research, which I used to collect grasping data from exteroceptive force sensors distributed along the gripper's fingers. Using the extracted data, I trained machine learning classifiers to perform object recognition from a known set of structural beam members and grasping Angles of Approach (AoAs). Due to similarities in structural beam member cross-sectional shapes and dimensions at particular AoAs, there are limitations to using grasping data from a single AoA for confident beam member recognition. To overcome this limitation, I have developed an information-based method for active tactile exploration of the structural members, which executes the minimum required number of grasps for confident object recognition. This method generalises well and can be used for any soft or compliant grasping mechanism, to recognise similar objects from a given target object set. My research publications relating to the topics discussed above can be found on IEEEXplore .