A robot’s neural network has many traits similar to the human brain, enabling continuous learning as it completes tasks in order to maximize efficiency. Through trial and error, both humans and robots are able to improve their abilities by repeating methods until they are able to find a process which works best.
Though this learning process is effective, it can be extremely time consuming for one robot to go through multiple trials before finding success. To make this process more efficient, Google has created a procedure allowing multiple robots to share experiences as they work, enabling one robot to teach another how to complete a task.
As each robot completes a task, information is uploaded to the server and the updated information is downloaded to other connected robots. Over a short period of time the connected robots learn from each other’s mistakes and complete the task flawlessly. To test the efficiency of this method, the Google Research team conducted three experiments to see how the robots dealt with learning simple tasks.
First, multiple robots who shared a network were tasked to open a door. After a few hours of blindly fumbling around, the robots discovered how to open the door and understood how their actions led up to their success. Next, multiple objects were placed in front of the robots, allowing them to move the objects around to understand the effect they had. As they pushed around the objects, the robots learned together the results that their actions had on each object. Last, the team used human guidance on a set of robots to physically move each robotic arm through the steps of opening a door. The robots were then tasked to copy the process and use trial and error to improve the method the human had set. Over time they not only found more efficient ways to open the door, but also learnt how to use a door handle.
Data monitoring, similar to the operations of Google’s neural networks, ensures that you receive the best in product and process control. Eclipse’s innovative methods, combined with expert in-house knowledge enable you to incorporate your analysis and tracking needs into a unique solution, offering you both a competitive and cost-efficient advantage.