Modular robots for sorting
McKenzie, Ross Malcolm
Current industrial sorting systems allow for low error, high throughput sorts with tightly constrained properties. These sorters, however, are often hardware limited to certain items and criteria. There is a need for more adaptive sorting systems for processes that involve a high throughput of heterogeneous items such as import testing by port authorities, warehouse sorting for online retailers, and sorting recycling. The variety of goods that need to be sorted in these applications mean that existing sorting systems are unsuitable, and the objects often need to be sorted by hand. In this work I detail my design and control of a modular, robotic sorting system using linear actuating robots to create both low-frequency vibrations and peristaltic waves for sorting. The adaptability of the system allows for multimodal sorting and can improve heterogeneous sorting systems. I have designed a novel modular robot called the Linbot. These Linbots are based on an actuator design utilising a voice coil for linear motion. I designed this actuator to be part of a modular robot by adding on-board computation, sensing and communication. I demonstrate the individual characteristics of these robots through a series of experiments in order to give a comprehensive overview of their abilities. By combining multiple Linbots in a collective I demonstrate their ability to move and sort objects using cooperative peristaltic motion. In order to allow for rapid optimisation of control schemes for Linbot collectives I required a peristaltic table simulator. I designed and implemented a peristaltic table simulator, called PeriSim, due to a lack of existing solutions. Controllers optimised in simulation often suffer from a reduction in performance when moved to a real-world system due to the inaccuracies in the simulation, this effect is called the reality gap. I used a method for reducing the reality gap called the radical envelope of noise hypothesis, whereby I only modelled the key aspects of peristalsis in PeriSim and then varied the underlying physics of the simulation between runs. I used PeriSim to optimise controllers in simulation that worked on a real-world system. I demonstrate the how the Linbots and PeriSim can be used to build and control an adaptive sorter. I built an adaptive sorter made of a 5x5 grid of Linbots with a soft sheet covering them. I demonstrate that the sorter can grade produce and move objects of varying shapes and sizes. My work can guide the future design of industrial adaptive sorting systems.