This Small Business Innovation Research (SBIR) Phase I project addresses the effort to design and prototype a new high-performance but low-cost integrated robotic manipulator comprising of the tight integration of mechanical, electrical, and computational subsystems. A significant part of the effort is to address the computational challenges involved in developing a new generation of intelligent robotic manipulators. Real-time general vision processing and dexterous planning, for example, are not feasible using conventional embedded processors. The consequence is that advanced robotic applications are only possible in environments where the robotic arm can outsource computationally expensive processes to more powerful computers. Low-power, High-Performance Computing (HPC) clusters will be used to extend what is currently possible in autonomous and semi-autonomous robotic manipulator systems. Therefore a research effort is to develop a scalable high-level intelligence framework applied to robotic manipulators, and to implement robust and real-time algorithms that take advantage of highly parallel computing environments. Application computations will integrate seamlessly across wireless and wired networks of heterogeneous robotic and computer systems. The goal is a highly capable, computationally scalable, low-cost intelligent robotic arm platform for research and light industry, which can easily be adapted to a variety of complex applications.
The broader impact/commercial potential of this project is to fill a market niche between the low-end robotic manipulators that have little commercial potential and the high-end robotic arms that are expensive and have high operating costs for setup and operation. High-performance computation integrated with advanced robotic manipulator systems are applicable for agile light industry and other desktop manipulator applications in unstructured environments. This will have a tremendous impact on the future of commercial robotics and make capable robotic systems affordable to small manufacturing businesses. Highly parallel computing power will greatly increase the range of applications and environments to which robots are suited. For example, an intelligent robotic manipulator can be used in agricultural applications or autonomously caring for plants in highly unstructured environments. The same manipulator could easily be adapted for use in a classroom setting; with a simple scripting interface students can experiment with advanced robotic control, allowing them to concentrate on discovering exciting new applications. This creates a broad market in academic and industrial settings, both for highly-capable, low-cost intelligent robotic manipulators for agile manufacturing and for lowpower HPC clusters applied to sensor network integration solutions in general.
The project researched the design and implementation of a scalable software architecture and parallel algorithms for real-time sensing and control applied to intelligent robotic manipulators. Computational and mechanical challenges, such as real-time general vision processing and dexterous trajectory planning, were addressed. Overcoming these challenges is key in developing the next generation of intelligent robotic manipulators. Additional research by RoadNarrows into embedded High-Performance Computing (HPC) clusters will make it possible for advanced robotic systems to be implemented in environments where computationally demanding tasks cannot be outsourced to more powerful computers, such as outdoor applications. RoadNarrows leveraged the company's current research into low-power HPC clusters and General Purpose Graphics Processing Units (GP/GPUs) to extend what is currently possible in autonomous and semi-autonomous robotic manipulator systems. RoadNarrows conceived a high-level software interface for a set of advanced control and sensing algorithms implemented in Message Passing Interface (MPI) and CUDA, NVIDIA's parallel computing architecture. The high-level software framework was partially implemented during Phase I as a set of proxy-server control modules which may be distributed over a network. Further work remains in developing robust parallel algorithms for sensing and control. In conjunction with the software effort, RoadNarrows continued the research and development of a family of low-cost commercial robot arms, known as HekaterosTM (see Figure 1), ideal for integrating low-power HPC clusters. The first arm developed was a 5 degree of freedom (DOF), lightweight, agile manipulator, designed for desktop mounting or on a durable mobile robot. The Hekateros manipulator was showcased in an autonomous robotic chess competition in August 2011, and performed very well. The arm consists of digitally manufactured ABS plastic frame components, Dynamixel actuators, color USB camera, and a fully embedded computing system, with the ability to lift and manipulate a 0.5kg object at a maximum reach of one meter. The RoadNarrows' robotic arm platform was developed to conform with current open software initiatives in robotics, such as Robot Operating System (ROS) and OpenRAVE. With open hardware and software interfaces, the resulting system is well suited for advanced research in robotics and for light industrial applications. An affordable robotic manipulation platform that is fully equipped with HPC capabilities and is preconfigured to support intelligent and sensor-rich applications will have a tremendous impact on the future of commercial robotics. Distributed and parallel computing will greatly increase the range of applications and environments to which robots are suited. This computing power will give the robot platform the ability to fuse a wide range of sensor data and thus constantly gather information and interact with its environment. For example, an intelligent robotic manipulator might be used effectively in an agricultural setting, such as autonomously caring for plants in an unstructured environment. The same manipulator could easily be adapted for use in a classroom setting; with a simple scripting interface students would be able to experiment with advanced control algorithms, allowing them to concentrate on discovering exciting new applications without the need to rely on a challenging and expensive IT infrastructure. An important goal of the project is to contribute to open standards, promote and use open-source and industry accepted standards in all aspects of the mechanical, electrical and software systems. All of these features create a broad market in academic and industrial settings for the project's highly-capable, low-cost line of intelligent robotic manipulators and the low-power HPC clusters applied to sensor network integration.