Data scientists at AI startups Neuromation and TRA Robotics have partnered to develop autonomous, intelligent and affordable robots that are able to not only take over daily household chores, but manage complex tasks in a variety of industries.
According to the companies, the prohibitive costs of AI development, the need for more high-powered computing, and a lack of the skills needed to train AI algorithms, are stumbling blocks to the development of these robots, and the partnership aims to address these.
Rodion Shishkov, CEO of TRA Robotics, says in order to transform a robot from dumb to artificially smart, you must first train its mechanical brain, the AI algorithms in its operating software, with immense amounts of data.
"Data is the lifeblood of the robotics industry. Unfortunately, the reams of data needed to train an AI algorithm to perform complex tasks in the home or in a particular industry are often in short supply. Depending on the complexity of its assigned tasks, the data needs of an algorithm may include millions of images to train a robot's optics and millions of sound files to train its hearing."
Synthetic data
Neuromation believes it has addressed this issue by creating vast pools of synthetic (artificial) data. Stored on a blockchain network, connected to the cloud, the data isn't real, but mimics real data closely enough that AI algorithms can use it to learn specific, complex tasks and finish their training.
"Synthetic data is Neuromation's strong suit. They'll help us create an army of smart machines and robots for the next evolution of the industry, Robotics 3.0," said Shishkov.
Over and above becoming a resource for synthetic data, the Neuromation platform will provide developers with the distributed computational power they will need to train deep learning algorithms. In addition, it is a reinforcement learning platform, which, inspired by behavioural psychology, allows AI algorithms to learn from their mistakes and perform their assigned tasks more reliably as a consequence.
Falling costs
Neuromation CEO Maxim Prasolov believes that by helping TRA and other developers train their AI-powered robots with synthetic data, reinforcement learning, and a powerful, distributed computing network, robot development costs will fall, resulting in a significant increase in the number of cost-effective and versatile robots brought to market for home services and industrial markets.
"Currently, a handful of huge tech companies own the industry, hiring the most talented AI researchers and buying all the AI startups," added Prasolov. 'We're flipping the script by creating a blockchain-based platform where an AI developer from different corners of the world can collaborate online, share or buy algorithms and synthetic data and then build tens of thousands of robots to manage complex tasks at home and work. It's an exciting time to be in the robot business."
Share