重要参考链接:
1. http://www.theconstructsim.com/ros-developers-live-class-develop-with-ros/
2. https://www.duckietown.org/research/AI-Driving-Olympics
We now have templates for ROS, PyTorch, and TensorFlow, as well as an agnostic template. Using these templates it takes about 5 minutes to make your first submission! We are also providing baseline implementations using our classical robotics pipeline, imitation learning with data from both simulation and real Duckietown logs, and reinforcement learning. There are more baselines and examples to get started with becoming available every day.
The schedule is as follows: (all tutorials will start at 9am EST)
Date | Organizers | Topic | |
---|---|---|---|
#1 | Fri. Nov 2 | Liam Paull Breandan Considine | Introduction to the AI Driving OlympicsGetting everything setupRunning the base templateMaking your first submission |
#2 | Mon. Nov 5 | Bhairav Mehta Liam Paull | Running the ROS templateUsing the Duckietown classical baseline |
#3 | Wed. Nov 7 | Manfred Diaz Julian Zilly | Running the TensorFlow templateImitation learning from Duckiebot logsImitation learning from simulator logs |
#4 | Fri. Nov 9 | Bhairav Mehta Florian Golemo | Running the PyTorch templateThe Reinforcement Learning baseline |
#5 | Mon. Nov 12 | Bhairav Mehta Sunil Mallya | Training in the cloud |
#6 | Wed. Nov 14 | Claudio Ruch Andrea Censi | The AMOD Challenge |
#7 | Fri. Nov 16 | TBD | Deploying on the Duckiebot |
#8 | Friday Nov 23 | All organizers | Ask us anything |
#9 | Friday Nov 30 | All organizers | Ask us anything |
#10 | Wed. Dec. 5 | All organizers | Ask us anything |
- Introduction to the AI Driving Olympics
- Getting everything setup
- Running the base template
- Making your first submission
#2Mon. Nov 5Bhairav Mehta Liam Paull
- Running the ROS template
- Using the Duckietown classical baseline
#3Wed. Nov 7Manfred Diaz Julian Zilly
- Running the TensorFlow template
- Imitation learning from Duckiebot logs
- Imitation learning from simulator logs
#4Fri. Nov 9Bhairav Mehta Florian Golemo
- Running the PyTorch template
- The Reinforcement Learning baseline
#5Mon. Nov 12Bhairav Mehta Sunil Mallya
- Training in the cloud
#6Wed. Nov 14Claudio Ruch Andrea Censi
- The AMOD Challenge
#7Fri. Nov 16TBD
- Deploying on the Duckiebot
#8Friday Nov 23All organizers
- Ask us anything
#9Friday Nov 30All organizers
- Ask us anything
#10Wed. Dec. 5All organizers
- Ask us anything
比赛通知:
人工智能驾驶奥运会(AI-DO),这是一场围绕AI自动驾驶汽车的新竞赛。比赛使用Duckietown平台17 ,它大量使用ROS。
AI-DO竞赛侧重于体现AI的难题,例如在复杂系统上部署学习,选择正确的学习过程模块化,以及在实际部署中学习模拟的挑战。
AI-DO 1与NIPS 2018一起举行。比赛将于10月1日至12月1日在云端和机器人中进行,并于12月7日在NIPS举行现场决赛。
AI-DO 2将在ICRA 2019举行,时间表将公布。
竞赛包括4个复杂性增加的挑战:1)空路上的道路跟踪; 2)道路跟随障碍; 3)与其他车辆在城市网络中进行点对点导航; 4)车队规划完整的自主移动点播系统。
参赛者将首先在模拟中发送他们的代码进行评估,然后在世界各地的Duckietown机器人中进行评估。不需要获取Duckiebot,但可以通过我们的kickstarter(bitly.com/DuckietownKickstarter 6 )获得。
人工智能驾驶奥运会由Duckietown基金会4 与6个学术机构合作推出:ETH苏黎世(瑞士),蒙特利尔大学(加拿大),清华大学(中国),国立交通大学(台湾),芝加哥丰田技术学院(美国)和佐治亚理工学院(美国)。
我们的两个行业赞助商和组织者是nuTonomy和亚马逊网络服务,除了财务支持外,还为项目提供最先进的基础设施。
要了解有关AI-DO的更多信息,请参阅http://AI-DO.duckietown.org 11 。
对应课程:
Duckietown项目致力于用机器人教授人工智能和机器学习。最近,他们推出了关于自动驾驶汽车的AI-Do驾驶奥运会比赛。
在本系列的三个Live Classes中,我们将看到如何使用模拟来编程de Duckiebots解决AI-Do竞赛和编程简单的自动驾驶汽车。
▸EP1| 2018年9月11日
在这个Live Class中我们将看到:
- 如何使用适当的ROS接口设置Duckietown和Duckiebots的完整Gazebo模拟
- 如何访问Duckiebots的传感器和执行器
- 如何为机器人创建一个简单的导航程序
▸EP2 | 2018年9月18日
在这个Live Class中我们将看到:
- 如何使用OpenAI创建一个遵循这些线的机器人。我们将使用强化学习(DeepQ学习)
▸EP3 | 2018年9月25日
在这个Live Class中我们将看到:
- 如何管理流量,避免其他Duckiebots和处理交叉点。我们将使用深度学习来训练机器人。
----
LIVE Class on how to develop with ROS. Every Tuesday at 18:00 CEST.
ROS Developers LIVE-Class #30
Duckietown AI Driving Olympics EP1 Sep 11th, 2018
Set Reminder ▸
ROS Developers LIVE-Class #31
Duckietown AI Driving Olympics EP2 Sep 18th, 2018
Set Reminder ▸
ROS Developers LIVE-Class #32
Duckietown AI Driving Olympics EP3 Sep 25th, 2018
Set Reminder ▸