【ICRA2019 AI Challenge Live Text Tutorial Session with Q&A】

Happyfs, Health1 + Health2 > Health1 + Health2. Whoever has more health wins

Q8. 场地周围的围栏是是什么颜色的 高反光白色？
A: In the mid-February, we will release detailed mechanical drawings about game field including projectile supplier and protective perimeter wall. And the colour of the protective perimeter will not influence the game.

arosendo, I don't think that is the case. They mentioned something about intial_health - remaining_health

Q9、关于硬件设备：针对去年比赛时购买的UWB在最终比赛场地时，标签标号设置问题导致UWB信息不可靠问题（貌似是大家采用相同标签标号导致），今年是否还会有这个问题？
A. This is an inevitable problem unless every robot out of battlefield turns off its UWB module.

Q10. 请问补给过程如何触发补给指令？官方有没有补给结束的指示？
Please refer to Q6

Q11. Should all the processing take place on the robot itself?
A. You can use your own computer out of the game field for computing and build a wireless link for data transmission. But the computer should be placed in the operator area.

Q12. Can we know the exact size of the Manifold2, worrying about the choice of other onboard computers?
A. Please refer to the robot size chart

请问比赛中什么样的冲撞会判犯规？只要碰到围挡或者障碍就算吗？双方机器人相互碰撞如何判罚？

Q13. When you mention "Visual Servo Control" for the Re-supply what does it mean? What is the precise mechanism that opens the small valve with 50 bullets? Being honest this doesn't look like require any kind of "vision" system.
In the mid-February, we will release detailed mechanical drawings about game field including projectile supplier and protective perimeter wall. And the wall of the projectile supplier will be attached with the certain marker.

Q2. 1. 关于模拟器的设计，我们发现实际上只需要不多的几个参数就可以描述整个游戏，比如用云台的俯仰角，底盘的角度和机器人的坐标就可以表示机器人的空间状态。所以我们打算这样来设计模拟器，就是先假设其他组的可以获得这些信息，然后我们来决策。这样简化了模拟器的设计，请问这样的设计会有什么问题吗？2.关于强化学习的，实际比赛的时候，我们不能直接获得敌方的比赛信息，血量，枪口热度之类的，但在模拟器里训练的时候可以获得，我们想在训练网络的时候输入实际可以获得信息，但奖励函数使用这些信息，也就是网络的输入与奖励函数的参数是不一样的。这样可以吗，会有什么问题吗？提前谢谢了
1. The abstraction level of observation space and action space will definitely have an impact on the performance. And an end-to-end system is not that stable in the real world. So you may need to train the network separately for the high noise model, or just substitute high noise model with the human-designed model. 2. A new firmware that would provide enemy information for training is in our future plan, but during the competition, you are not able to know enemy information. So reward shaping may be tough work.