Lange introduced that machine learning is very important for game development. It can make the game development process easier and make the game more attractive. The developer can program without every program. The interaction between the game and the player can make the system Self study. Just as people perceive and react to the environment. ML-Agents can train these systems in the same way.
Lange presents a Demo, futuristic racing game where the human being on the left is the human and the machine learning Agent on the right. In the beginning, the Agent didn’t make a good turn and it was easy to crash, but it slowly learned the human operation. After about 25 minutes of training, you can see that it may still be a bit unstable, but it will not crash anymore. After continuous training, Agent can program very good players.
Then, in the specific game scenario, there can be the following applications: Machine learning can be applied in several ways. The first is the NPC that creates NPCs and has multiple behaviors. The NPCs have learned a lot of human behaviors, which will make interaction with humans more natural. The second is the game itself. AI can learn to optimize the fun of the player, not optimize for the fun of the developer. For the player, there will be more personalized and customized things. The third is an entirely different area where we use machine learning to test the game before the game is released. Make sure that when you use Agent to play games instead of human players, you can see if the game goes smoothly.
In addition, if there are few human players in the game, Agent can be used instead of human players. The last aspect is match making. Use machine learning to find the right players and match them together to play the game. You can let the machine learning system know how to optimize the game time, match players, and how to maximize the use of game time.
Since more than two years ago, Unity has begun its transformation from a pure engine provider to a network + cloud value-added service. Danny Lange is the driving force behind Unity’s development in the AI field. Prior to joining Unity, Lange served as head of Uber machine learning. Earlier, Lange was responsible for the development of machine learning products for Amazon and Microsoft.
In addition to games, ML-Agents still have many application scenarios in the field of autopilot and robotics. Lange introduced that autopilot is a big application area for ML-Agents because it does not need to drive on real roads like Uber. Machine learning simulation can avoid accidents caused by real road tests. Another area is robotics. With enhanced learning, you can train robots in a virtual environment and you can quickly complete hundreds of thousands or millions of trainings. The trained model can be applied to real robots.
In addition, ML-Agents can also be used in architectural design, such as how to rationally design channels, people flow, and use machine learning to simulate roadmaps in buildings.