The first use case is about autonomous driving. We use the concept of horizontal federated machine learning to accelerate the training of reinforcement model. There are four agents here, three real cars and one in the simulator.
These two short videos show the result of a federated reinforcement model. You can see from these videos that both the real car and the cat in the simulator perform well after serveral iterations.
The second use case is about applying federated machine learning for warehouse monitor. In this case, we need to build a model to recognize the goods type and volume. But a single warehouse can not provide enough data and upload all the monitor video is costy. So we use federated machine learning to solve this problem.
This video shows how the model works. When the staff withdraws goods, we will take a picture of the goods and use the federated model to recognize the type and another federated model is used to calculate the good's volume. we also provide a virtual electric fence to capture the unexpected entrance of a certain area. You can see that the staff with a pink t-shirt is captured when he enters the zone that surrounds by the red lines.