How robots understand human feelings

After listening to Professor Ren's speech today, he was inspired and his ideas became clearer. The professor's speech was so condensed that everything was up to nothing. We need to record our own digestion.
Robot understanding of human emotions can be roughly divided into the following three steps:

1. Identify:
Recognition includes the recognition of human expressions and sounds. This includes the judgment of context and natural language processing. This is divided into two pieces:
A. Facial recognition: Facial expression recognition can be broadly divided into several common patterns: happy, happy, excited, angry ... Each of these modes can be converted to a different pattern, but the probability of conversion is different, as shown in the following figure: Professor here has a higher demand for visual recognition of robots, not just visual single frame image recognition, but continuous image processing, where the human expression of continuous processing is called "emotional calculation." That is, the image context of the association, here to our visual platform to inspire, the robot to accurately identify the human face needs not only the current static picture, usually need a set of pictures or video sets;
B. Tone recognition: Tone can also be divided into several common patterns, but the tone of the recognition is much more difficult, Professor here is the introduction of Chinese natural language processing technology, I can add here, hit also is doing modal recognition of the topic, this can be cooperative, not only from the natural language processing perspective to analyze semantics, but also to let the robot really listen to Understand

2. Understanding:
Understanding includes the robot's comprehensive analysis of human behavior, and accurately determine the human emotional state. Professor here has proposed a "abundant heart engineering" model based on large data, the literal meaning is to enrich the robot's mind, from the technical point of view is to collect a lot of human psychological text description samples, and ultimately is a pattern recognition based on Chinese characters.
As more and more text information is collected, the accuracy of robot recognition will become higher.

3. Feedback:
After understanding the human mind, the next step is to respond, and the professor here has no specifics on how to respond. He showed us the prototype of his humanoid robot, which is more acceptable from the angle of human emotion, and it is very shocking to see the incarnation of Professor Ren. It's just a trick. But the reaction is slightly slower, as shown in the following picture: After reading, I still have a few puzzles:
A. How to solve the robot's reaction speed.
B. How to solve the robot's energy endurance capacity.
C. How to solve the safe use of robots. (if it's because of an unusual crime, it's very controversial.) ）

Finally attach the professor's speech Youku playback address:
http://v.youku.com/v_show/id_XNzQyNDczODcy.html?f=22783846&sf=10102