AI-powered metaverse: Training robots
For instance, Siemens’ SIMATIC robot pick AI extends this vision of adaptability by transforming industrial robots, which were once limited to rigid and repetitive tasks, into complex machines. The AI is trained on synthetic data, which are virtual simulations of materials, shapes and environments. It prepares robots for unpredictable tasks like picking unknown objects from chaotic bins with 98% accuracy. The system improves through feedback from real-world situations when mistakes are made. This is not a single-robot solution. Software updates scale across entire fleets, upgrading robots to work more flexibly and meet the rising demand for adaptive production.
Another example is the robotics firm ANYbotics, which generates 3D models of industrial environments that function as digital twins of real environments. The integration of operational data such as pressure, temperature and flow rates is used to create virtual replicas that are similar to physical facilities. A power plant can, for instance, use site plans to create simulations of the inspection tasks that robots will perform at its facility. This speeds the robots’ training and deployment, allowing them to perform successfully with minimal on-site setup.
Simulation also allows for the near-costless multiplication of robots for training. In simulation, we create thousands of virtual robotics to perform tasks and optimize their behaviour. This allows us to accelerate training time and share knowledge between robots,” says Peter Fankhauser, CEO and co-founder of ANYbotics.
Because robots need to understand their environment regardless of orientation or lighting, ANYbotics and partner Digica created a method of generating thousands of synthetic images for robot training. By removing the painstaking work of collecting huge numbers of real images from the shop floor, the time needed to teach robots what they need to know is drastically reduced.
Similarly, Siemens leverages synthetic data to generate simulated environments to train and validate AI models digitally before deployment into physical products. By using synthetic data we can create different variations in lighting, object orientation and other factors, ensuring that the AI is able to adapt well under various conditions. We simulate everything, from lighting and shadows to how pieces are oriented. This allows the model to train under diverse scenarios, improving its ability to adapt and respond accurately in the real world.”
Digital twins and synthetic data have proven powerful antidotes to data scarcity and costly robot training. Robots can prepare themselves quickly and cheaply in artificial environments for a wide variety of visual possibilities. De Paola says that they validate their models in a simulated environment prior to deploying them. This technology can have a far-reaching impact beyond the initial robot training. The well-trained robot at work
With simulation and AI powering a brand new era of robot training, organisations will reap the rewards. The use of digital twins allows companies to deploy advanced robots in a much shorter time frame. AI-powered vision system can also be used to adapt product lines to meet changing market needs.
