Although NVIDIA has proposed methods including the Isaac virtualization platform for robot training, which allows robots to complete a large number of motion instructions in a short period of time through virtualization, if robots are to learn more delicate movements, they still need human correction.Through virtual reality, allowing trainers to teach the correct movement details "hand in hand" through the robot's perspective.
The reason why we hope that robots can perform more delicate movements is naturally because we expect robots to be able to correctly hold the door handle and open the door, or grab and move the cup. In the future, they may even be able to further complete more applications such as home care or dangerous places that require more delicate operations.
Training through virtual reality will allow the trainer to correct the robot's incorrect operation in real time through the robot's perspective, while also allowing the robot to learn the correct movement details more quickly and more easily learn how to deal with different situations.
In addition to training through virtual reality, Toyota's research center's training design allows robots to learn object-to-object interactions through computer vision, rather than relying on positional memory to determine interaction patterns. This prevents objects from changing position and causing the robot to become confused. Furthermore, once training is complete for one robot, the system can synchronize training across the network, allowing all robots to learn the same skills.
Toyota Research Institute stated that this training is not a specific product concept design, but rather a prototype design of a training model, allowing more robots to learn different application methods through different training scenarios, thereby completing more tasks that originally required precise human operation.
