Google is thinking about using humans to remotely train robots to perform complex or unfamiliar tasks on a production line.
Robots are usually pre-programmed for very specific tasks. For instance, if a robot has to repeatedly apply a fastener to a specific component on an assembly line, the instructions would be: grasp a component from one fixed location, apply it to another fixed location, and repeat.
However, in scenarios where there are a lot of components, or some new components that a robot hasn’t seen before, a significant engineering effort is usually required in pre-programming the robot to make it fully productive.
To quickly train robots in these circumstances, Google is thinking about having people remotely control a robot to perform a task on the assembly line. This could be through people using VR headsets to share the same ‘view’ as a robot, and then controlling the robot’s hands with VR gloves or hand controllers.
Based on this ‘human input’. a machine learning model can then be trained in order to predict what the robot needs to do in order to complete a task successfully. In turn, the robot will quickly be able to work autonomously, or at the very least, with minor human intervention to ensure full productivity on the assembly line.
This filing is interesting for a whole host of reasons. The one that stands out to me is that we have an example of how VR, gaming and remote work can come together. One opportunity I’ve been thinking about recently is the idea of using game-like interfaces to make training machine learning models fun, quick and cheap. These games could be built to be extremely entertaining, financially rewarding, and have real-world utility.
One future prediction is that the worlds of enterprise, gaming and AI will converge in fascinating ways, and this filing is a hint of this possibility.