Duckietown

🎓 Planning on teaching a robot autonomy undergraduate or graduate level class? Check out the #Duckietown Instructor Manual: https://hubs.la/Q02MRHFK0 📚 It provides an overview of the Duckietown resources available for teaching, as well as tips and tricks to run a successful class. 🔔 Follow us at Duckietown 🖥️ Visit our website - https://hubs.la/Q02MRGxD0 #RobotAutonomy #STEMEducation #TeachingRobotics #HigherEd

🧠Interested in learning more about how robots work and take decisions autonomously?🤖 Join the “Self-Driving Cars with Duckietown” online course for free at: https://hubs.la/Q02Mlq6t0 🔔Follow us at Duckietown 🖥️Visit our website - https://hubs.la/Q02Mlh-D0 #Duckietown #Duckiebot #Robotics #AI #engineeringeducation

Wouldn't it be nice to be able to train multi-agents policies in simulation, and then deploy them in the real world worry free for safety-critical systems? Eduardo Candela, Leandro Parada, Luis Marques, Tiberiu Andrei Georgescu, Yiannis Demiris, Panagiotis Angeloudis from Imperial College London created a robust pipeline for training multi-agent RL policies that is not only effective in simulation but also highly adaptable to real-world conditions. Check out the results! https://hubs.la/Q02M1KfR0 🔔 Follow us at Duckietown 🖥️ Visit our website - https://hubs.la/Q02M1N1M0 #Duckietown #Duckiebot #Robotics #AI #Sim2Real

Learn more about Kevin Smith and his teaching experience at MassRobotics. Read the full interview here: https://hubs.la/Q02LxCz50

Can you make a Duckiebot autonomously navigate from a starting tile to any (compliant) arrival tile in a Duckietown, and stop there? Johannes Boghaert implemented a planner based on Dijkstra’s algorithm and leveraged Duckietown road infrastructure to achieve global localization, and have Duckiebots reach their destinations safely. Learn more and start reproducing: https://lnkd.in/d2D96zSy 🖥️Visit our website - https://hubs.la/Q02L5jxq0 #Duckietown #Duckiebot #Robotics #AI #AutonomousNavigation

Perception is a tricky thing, and when it comes to robots having to operate autonomously in the environment, it is even trickier. Are you planning on teaching perception, planning, control or other topics related to robot autonomy? Take a look at Duckietown: https://hubs.la/Q02KxVDK0

How does visual domain randomization help improve sim-to-real transfer, or the real-world performance of robotic agents trained in simulation? Visual domain randomization involves changing the appearance specifications of a simulated environment during different iterations of training of an agent so that it is exposed to a distribution of scenarios before being deployed in the real world. András Béres and Bálint Gyires-Tóth of Budapest University of Technology and Economics dive deep into this topic in their work: “Enhancing Visual Domain Randomization with Real Images for Sim-to-Real Transfer", proposing to incorporate real images with visual domain randomization in reinforcement learning training to improve sim-to-real transfer performance. They used the Duckietown Gym Simulator to test their method, and the results were very interesting! Read more about their work: https://hubs.la/Q02JbqB60

How do autonomous vehicles detect objects in their surroundings, like duckies-pedestrians-obstacles on the road? Collision-free travels are a desirable asset when riding an autonomous vehicle. Robust detection of a broad array of objects in varying lighting conditions is an important component of any obstacle avoidance pipeline. Former Duckietown students Maximilian Stölzle and Stefan Lionar, implemented a YOLO-based object detector that performs amidst varying lighting conditions, demonstrating: - Lane Following with object detection - Duckiebots autonomously stop upon detecting obstacles like Duckies or other Duckiebots. Read more and reproduce their work: https://hubs.la/Q02HXmVM0 #robotics #computervision #selfdrivingcars #machinelearning #safety #duckietown #yolo #obstacledetection #obstacleavoidance

How to better understand how RL trained ML agents take decisions? This paper explores reward consistency for better interpretability in reinforcement learning. Qisen Yang, Huanqian Wang, Mukun Tong, Wenjie Shi, Gao Huang, and Shiji Song from Tsinghua University approach the black-box nature of deep reinforcement learning (RL) and its limitations in real-world applications in their paper titled “Leveraging Reward Consistency for Interpretable Feature Discovery in Reinforcement Learning”. They tested their “RL-in-RL” approach on Atari 2600 games and Duckietown Gym. Check out the results: https://hubs.la/Q02HpYWK0

🌙 Autonomous driving is hard, and driving at night is even harder! While it might seem like a trivial difference, environmental lighting conditions play a huge role in the performance and robustness of autonomy pipelines, especially lighting conditions when vision is used as main sensing modality. With the flip of a switch, Duckietown offers the opportunity to tackle hard challenges. Start teaching and learning robot autonomy now: https://hubs.la/Q02HcDdj0 ✨ #diy #engineeringeducation #robotics #engineering #AI #highereducation #selfdrivingcars #computervision #robotautonomy