【ICRA2019 AI Challenge Live Text Tutorial Session with Q&A】
training spl
8885
0
128
2019-01-30
新闻：ICRA 人工智能挑战赛在线文本教程将于本周四举行！
北京时间（UTC+8）2019年1月31日上午10点，RoboMaster研发部的Charles Yang将为您带来直播文字直播：ICRA挑战赛准备技巧。请方便地观看本文中的实时文本并在会议期间提出问题。
如果您对如何更好地开发机器人有任何疑问，请随时在下面留下您的问题，主持人将在直播中整理这些问题并做出相应解答。
议程：
上午 10:00 - 10:10 - 对提交的团队技术提案进行一般性讨论。
上午 10:10 - 10:20 - 准备 ICRA 挑战的提示。
上午 10:20 - 11:00 - RoboMaster AI 团队实时答疑。
主持人：@Training Spl
特邀发言人：
- AI挑战赛总工程师 - Charles Yang - 论坛ID：charles.yang
- AI挑战赛工程师 - Vic Wen - 论坛ID：vvwen
- AI挑战赛工程师 - Kevin Li - 论坛ID：KevinLAD
在直播过程中，主持人和演讲嘉宾只会输入英文字母，无论用哪种语言来描述问题。
-------------------------------------------------- -------------------------------------------------- -----
直播内容：
在收到各参赛队的提案后，总结一下，我们发现大部分参赛队已经对比赛场景的机器人系统设计有了很好的掌握。但仍存在一些普遍性问题，可能对备赛工作造成负面影响：
1. 以下是您必须在游戏中实现的几个基本模块：
a) 定位
b) 运动规划和控制
c) 具有弹丸轨迹预测的视觉（装甲检测）万向伺服控制
d) 弹丸供应（从导航到视觉伺服控制）
e) 决策系统
b) 运动规划和控制
c) 具有弹丸轨迹预测的视觉（装甲检测）万向伺服控制
d) 弹丸供应（从导航到视觉伺服控制）
e) 决策系统
此外，敌方机器人检测、游戏场地分割也可以应用于机器人。
2. 更加关注NAVIGATION系统的定位和避障运动规划。它是机器人的基本元件。仅知道如何运行开源包并不能保证你的机器人的性能。碰撞处罚是严厉的，没有任何让步。
3. 确保您的解决方案实用且可行，并更多地考虑工程。
For example,
a) if you use model-based filtering algorithm, please build your sensor model accurately enough or your algorithm robust enough in the game scenario, especially for UWB and IMU.
b) If you use learning based algorithm, take care of the computing source for onboard PC so that all algorithms can process in acceptable and steady frequencies at the same time.
c) If your algorithm highly depends on visual features. Think more about the light condition and possible accidents for enemy robots in the game scenario.
d) If you want to set up a wireless link for communication between robots, take the transmission quality and delay into consideration.
a) if you use model-based filtering algorithm, please build your sensor model accurately enough or your algorithm robust enough in the game scenario, especially for UWB and IMU.
b) If you use learning based algorithm, take care of the computing source for onboard PC so that all algorithms can process in acceptable and steady frequencies at the same time.
c) If your algorithm highly depends on visual features. Think more about the light condition and possible accidents for enemy robots in the game scenario.
d) If you want to set up a wireless link for communication between robots, take the transmission quality and delay into consideration.
4. A stable system is a time-consuming work in the competition preparation period. It should take more than half of your preparation period to TEST instead of DESIGN. As the whole system is a combination of different modules, the buckets effect will have an obvious influence on the system performance, especially for visual servo control. Accurately targeting and hitting the armour of enemy robots needs to fine-tune detection, projectile trajectory prediction and gimbal servo control. In addition, decision making in the simulation is quite different from that in the real game scenario. Partially observable decision process needs testing to find the appropriate action space.
Now it is time for you to implement your solution on the real robot platform. Here is some advice about tuning the robot:
1. When you received the robot, firstly read the 《ICRA 2019 DJI RoboMaster AI Robot User Manual》. Here is the link to download it:
ICRA 2019 RoboMaster AI Challenge Important Documents:
https://www.robomaster.com/en-US ... ype=announcementSub
https://www.robomaster.com/en-US ... ype=announcementSub
a) Check your circuit topology to make sure each module is connected in the right way.
b) You must know how to use the remote controller to control your robot, including chassis and gimbal motion, turning on the friction wheel and launching the projectiles under manual mode.
b) You must know how to use the remote controller to control your robot, including chassis and gimbal motion, turning on the friction wheel and launching the projectiles under manual mode.
2. Get onboard PC ready for powering on with the proper voltage transformer and connected with MCU through a USB cable.
3. Test the robot driver on the onboard PC, make sure the robot is placed in the OPEN and SAFE area:
a) According to the tutorial, install the software dependencies, compile the robots packages and configure the udev rules on your onboard PC.
b) Stay on manual mode for remote controller, run the roborts_base_node with Rviz under ROS. Use the remote controller to control the robot, and check if you can get the correct odometry and transform topics.
b) Stay on manual mode for remote controller, run the roborts_base_node with Rviz under ROS. Use the remote controller to control the robot, and check if you can get the correct odometry and transform topics.
c) If everything goes well, publish topics for gimbal or chassis control and switch remote controller to autonomous mode. Check if the robot moves correctly.
Here are some new information for the competition:
1. 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.
2. RoboMaster organization will not provide a standard simulator for the game this year.
3. The communication between robots through the referee system is still under development and is supposed to finish on Mid-March. So teams had better build your own wireless communication link.
4. We will be glad to hear from teams for the definition of robot illegal behaviours that would take advantage of loopholes in the rule but no more help the inspiration and development of technology. The rules will forbid such behaviours by common consent.
2. RoboMaster organization will not provide a standard simulator for the game this year.
3. The communication between robots through the referee system is still under development and is supposed to finish on Mid-March. So teams had better build your own wireless communication link.
4. We will be glad to hear from teams for the definition of robot illegal behaviours that would take advantage of loopholes in the rule but no more help the inspiration and development of technology. The rules will forbid such behaviours by common consent.
-----------------------------------------------------------------------------------------------------------------------------
Q&A
Q1. 关于补给站的问题：1. 能知道补给站的出弹速度吗？2. 手册里说：一分钟能接受任意一个机器人的两次补弹指令，指的是一个机器人两次，一共四次吗，还是一共两次？谢谢啦
A：1. Yes, currently the time of supply for 50 projectiles is about 3 seconds. 2. 2 times in total in 1 minute no matter how many robots get supplied.
Q2. 关于比赛规则：在比赛规则的获胜条件中有一条：“一局比赛时间耗尽时，双方机器人尚且存活，伤害高的一方获胜”。伤害高的一方获胜是指剩余血量高的一方获胜？还是命中次数多的一方获胜？伤害来源不只是射击伤害，还有其他的扣血机制。这里伤害是指（初始血量-对方剩余血量），还是只指射击命中伤害？
A：The initial value minus the remaining blood is the damage caused to the enemy.
Q3. 关于比赛规则：在淘汰赛中，上局比赛结束到下局比赛开始的时间间隔是多少？
A：After one game ends, it will jump to Five-Minute Setup Period of next game.
Q4. 关于比赛规则：补给区的触发条件？补给时长？如果由补给装置发生卡弹造成子弹补给不足如何解决？
A: After the robot sends the command to the referee system, the supply station will start replenishing.
Q5. Would mind to give some tips about when can we purchase manifold2 and referee system? And, how much are they?
A: Manifold2 will be on sales roughly in the middle of Feb. The referee system will be ready at the early March. Please pay attention to our latest announcement.
Q6. 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.Q7. 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.
Q8. 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.
Q9. 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
Q10. 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.
Q11. 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.
Q12. Is it possible to know the specific about the given IMU on the robot? Thanks
A. Development board type A： MPU6500+IST8310
Single axis gyro module: ADXRS620
Q13. "A：The initial value minus the remaining blood is the damage caused to the enemy."
Therefore, if we have one robot against two, we have double health?
A. No
Q14. 补弹指令具体指什么呢？
A. It is a protocol command sending to Referee system (from STM32 MCU to Referee system). But in the later robots version, we will provide it in the ROS sdk also. So no worry about this question.
Q15. 请问比赛中什么样的冲撞会判犯规？只要折叠围挡或者障碍么？双方机器人相互碰撞如何判罚？
A.参考规则手册3.5.3.1攻击与竞争（碰撞罚分）Q16.您好，我们购买了官方车，但是买来的机器人云台是不可控的，请问车上的云台板是完全留待自己开发的吗，还有和github上的官方队列程序是基础板的程序还是云台板的程序？非常感谢！！A. 请确保您的遥控器操作模式正确（请参阅AI Robot用户手册）。并且STM32程序是开源的（RoboRTS-Firmware：https://github.com/RoboMaster/RoboRTS-Firmware icra2019_dev分支）。板载ROS程序开源（RoboRTS： https: //github.com/RoboMaster/RoboRTS ros分支）Q17. 关于“弹丸储存器”的问题。我知道我们应该扩展它，但最大的问题是：它可以扩大多少有限制吗？我假设只要我们遵守 600x600x500 的限制就可以了？A. 只需满足裁判系统检查规范，参见规则手册附录四-赛前检查表Q18。在目前的ROS API中，我们如何控制发射动作呢？主题是什么？A. /cmd_shoot (roborts_msgs/ShootCmd)弹丸发射服务命令（包括模式、频率、数量） 参考https://robomaster.github.io/Rob ... k_docs/roborts_base（英文版几天后发布）Q19.最终50发弹丸数量如何计算？是否会根据上供弹无法完全交付弹丸仓库内全部弹丸而考虑实际装弹量？清理弹仓时，弹管内的弹丸不会被清理。
文章标签
请问这篇文章对你有用吗？
【ICRA2019 AI Challenge Live Text Tutorial Session with Q&A】