Robotics Academy: Learn Robotics, AI and Computer Vision
- 提供者: JdeRobot (Universidad Rey Juan Carlos)
- 講師: José M. Cañas and the JdeRobot community
- 程度: intermediate
- 形式: Self-paced exercises
- 時長: Self-paced (43 exercises, roughly 1-3 hours each)
- 類別: Robotics
- 語言: English
- 價格: 免費
- 標籤: robotics, computer vision, ROS 2, Gazebo, autonomous driving, drones, navigation, SLAM, python, open source
一套免費、採用 GPL-3.0 授權的 43 個機器人程式設計練習,來自 Universidad Rey Juan Carlos 的機器人研究團隊 JdeRobot。每個練習都會給你一台模擬機器人和一個空函式:你要撰寫 Python 程式,讓 F1 賽車沿線行駛、讓無人機逃出機庫、讓掃地機器人清掃整個房間,或讓 TurtleBot 找到走失的人。所有內容都在預先準備好的 Docker 映像檔中的 ROS 2 Humble 與 Gazebo 上執行,因此模擬器、OpenCV 與 PyTorch 都已安裝完成,你可以直接在瀏覽器中編輯、執行與除錯。同樣的練習也可以在 JdeRobot 的託管版本 Unibotics 上完成,完全不需要安裝。
為何值得修讀
Most robotics courses teach you to read about robots. This one only ever asks you to program one. There are no video lectures to sit through and no toy notebook problems — each exercise is a Gazebo world, a real ROS 2 stack, and a scoring condition you either meet or do not, which is much closer to the experience of actually making a robot work. It is also genuinely open: the exercises, the simulation worlds, the web IDE and the Docker images are all in a GPL-3.0 repository that has been maintained since 2015 and is still committed to daily, so you can read how any exercise is built, fix it, or add your own. Two auxiliary modules, HAL.py for sensors and actuators and WebGUI.py for browser visualisation, keep the boilerplate out of the way so the algorithm is the thing you write.
你將學到
- Write robot control algorithms in Python against a live ROS 2 and Gazebo simulation
- Reactive navigation: line following, virtual force field obstacle avoidance and the dynamic window approach
- Global path planning with visibility graphs, gradient fields and RRT
- Probabilistic localisation with particle filters, using laser or vision
- Robot vision: colour segmentation, 3D reconstruction from a stereo pair, visual odometry and marker-based localisation
- Deep learning in the loop: image classification, object detection and end-to-end visual driving
- Aerial robotics: position control, visual landing, drone gymkhana and cat-and-mouse pursuit
- Service and industrial tasks: warehouse logistics, pick and place, mobile manipulation and package delivery
課程大綱
There is no fixed order: the 43 exercises are grouped into Autonomous Driving, Service Robots, Drones, Computer Vision and Industrial Robots, and you pick what you need. Each one ships as three parts — the Gazebo configuration, a browser page with the code editor and visualisation, and documentation carrying the theory and hints for that specific problem. Offline you launch RADI, the RoboticsAcademy Docker image, which contains ROS 2 Humble, Gazebo, OpenCV and PyTorch preinstalled, then connect a browser to the Django server running inside it. Online you skip all of that and use Unibotics. Difficulty ranges widely: follow line is an afternoon, while Monte Carlo localisation or end-to-end visual control are closer to a coursework project.
核心主題
- Autonomous driving: follow line, follow road, autoparking, road junction, qualifying lap
- Mobile robot navigation: obstacle avoidance, global navigation, laser mapping, vacuum coverage
- Localisation: Monte Carlo with laser or vision, marker-based visual localisation
- Computer vision in robotics: 3D reconstruction, optical flow teleoperation, visual odometry
- Deep learning: image classifier, visual object detection, end-to-end visual control
- Drones: gymkhana, hangar escape, visual lander, power tower inspection, cat and mouse
- Service robotics: rescue people, follow person, Amazon warehouse, package delivery
- Industrial robotics: pick and place, mobile manipulation
先備知識
- Comfortable programming in Python
- Basic linear algebra and geometry — frames, transforms, angles
- Docker and a Linux machine for the offline route, or nothing at all if you use Unibotics in the browser
- No robot hardware required: every exercise runs in Gazebo
適用對象
- Students taking a robotics or computer vision course who want practice rather than theory
- Engineers moving into ROS 2 who learn better from a working simulation than from documentation
- Educators looking for ready-made robotics assignments with a browser-based submission route
- Self-taught roboticists with no hardware who still want something that behaves like a real robot