According to foreign media reports, anyone who has considered expanding business or building a network should be familiar with the so-called "network effect." For example, the more buyers and sellers are using eBay and other platforms, the more useful it will be. The data network effect is a dynamic effect. With the increase in the use of such services, it actually helps to improve the service itself. For example, with the increase of training data received, machine learning models usually become more accurate. Driverless cars and other intelligent robots rely on sensors to generate more and more highly variable data. These data are used to build better AI models, and robots rely on these models to make real-time decisions and navigate in real-world environments. As the core of today's intelligent robots, the fusion of sensors and AIs is generating a benign feedback loop, or what we call the "robot network effect." We are currently at the edge of the "robot network effect" critical point, which will greatly accelerate the development of robotics technology. Rapid development of AI In order to understand why robots are the next frontier for AI, we need to step back and first understand how AI evolved. Machine intelligence systems that have been developed in recent years can use a large amount of data that did not exist in the mid-1990s, when the Internet was still in its infancy. Advances in storage and computing technologies have made it possible to quickly and inexpensively store and process large amounts of data. However, the improvement of these projects cannot explain the reasons for the rapid development of AI. Open source machine learning libraries and frameworks play a quiet but equally important role. Fifteen years ago, when the scientific computing framework Torch was released under the open source license of BSD, it contained many algorithms commonly used by data scientists, such as deep learning, multi-layer perceptrons, support vector machines, and K-nearest neighbors. Wait. Recently, open source projects such as TensorFlow and PyTorch have made valuable contributions to this shared knowledge base, helping software engineers with different backgrounds to develop new models and applications. Domain experts need a lot of data to create and train these models. Large enterprises have great advantages because they can use existing data network effects. Sensor data and processing capabilities Lidar sensors have existed since the early 1960s and have been used in geography, archaeology, forestry, atmospheric research, defense, and other fields. In recent years, laser radar has become the sensor of choice for unmanned navigation. The lidar sensor on Google's unmanned vehicles produces 750MB of data per second, and eight automotive computer vision cameras can generate another 1.8GB of data per second. All of this data must be processed in real time, but centralized computing (in the cloud) is not fast enough to handle real-time, high-speed operations. To solve this bottleneck, we began to develop edge computing. On robots, we use on-board calculations. The solution for most current driverless cars is to use two car "boxes," each equipped with an Intel Xeon E5 CPU and four to eight Nvidia K80 GPU accelerators. At the highest performance, this consumes more than 5000 watts of electricity. Recent hardware innovations such as Nvidia's new driver, PXPegasus, can support 320 trillion calculations per second and begin to address this bottleneck more effectively. Breakthrough in AI Our ability to simultaneously process sensor data and fuse a variety of data will continue to drive the evolution of intelligent robots. In order for this sensor fusion to occur in real time, we need to place machine learning and deep learning models in the edge calculations. Of course, decentralized AI increases the demand for distributed processors. Fortunately, machine learning and deep learning calculations are becoming more efficient. For example, the cost of Graphcore's intelligent processing unit (IPU) and Google's tensor processing unit (TPU) is declining and accelerating the performance of neural networks on a large scale. In other areas, IBM is developing neuromorphic chips that mimic brain anatomy. Its prototype uses 1 million neurons and each neuron has 256 synapses. This system is particularly suitable for interpreting sensory data because it is designed to mimic the human brain's interpretation and analysis of sensory data. All of these data from the sensors show that we are on the verge of “robot network effects†and that this shift will have a huge impact on AI, robotics, and their various applications. New data world "Robot network effects" will enable new technologies and machines not only to process more data faster, but also to expand the types of data. The new sensors will be able to detect and capture data that we may not have considered at all, because of the limited human perception. Machines and smart devices will contribute rich data to cloud and neighboring agents, inform decisions, enhance coordination, and play an important role in continuous model improvement. These advancements are much faster than many people realize. For example, Aromyx uses receptors and advanced machine learning models to build sensor systems and a platform for digitally capturing, indexing, and searching for odors and taste data. The company’s EssenceChip is a disposable sensor that can output the same biochemical signal that the human nose or tongue emits when it smells or tastes food or drink. OpenBionics is developing a robotic prosthesis that relies on tactile data collected from sensors embedded in the arm cover to control the palm and finger movements. This non-invasive design uses a machine learning model to translate the fine muscle tension on the electrodes into the complex motion response of the bionic hand. Sensor data will help drive the development of AI. The AI ​​system will also expand our ability to process data and discover the creative uses of this data. In other respects, this will inspire new morphological factors for robots and enable the collection of a wider range of data models. As we "see" our ability to evolve in new ways, the everyday world is quickly becoming the next great frontier for technology discovery. German type cable reel Non-rewirable, 230V~, IP20, Class I, with a thermal cut-out, with a non-rewirable plug with cord H05VV-F 3G1,0-1,5mm 2 , Germany Cable Reel,Cable Reel,Automatic Cable Reel,Cable Spool Reel CIXI KYFEN ELECTRONICS CO.,LTD, , https://www.kyfengroup.com
with 4-way socket outlet with or without shutters.