Duckietown

Guillaume Gagné-Labelle, Gabriel Sasseville, and Nicolas Bosteels from the Mila Institute in Montreal, Canada, explore how computation delays affect reinforcement learning performance in Duckiematrix autonomous driving tasks.

Using Soft Actor-Critic models in Duckietown simulation, the team compares classical RL policies with action-conditioned Real-Time RL approaches under fixed and variable latency conditions, evaluating their impact on reward stability, policy performance, and episode length.

The project highlights how real-time constraints can influence autonomous driving behavior and policy robustness in embodied AI systems.

Learn more: https://hubs.la/Q04hNLqQ0

Don’t miss the , taking place April 28-30 in Rome, hosted by Università La Sapienza of Rome (Marco Polo building) and promoted by Fondazione Mondo Digitale.

The Rome Cup brings together schools, universities, research centers, companies, and institutions to explore how robotics and AI can be meaningfully integrated into education and career pathways.

This year’s theme is “augmented intelligence”, focusing on how humans and artificial systems work together in real contexts.

Duckietown will be represented by Jacopo Tani, and will be part of a panel of judges to assign an award to the best project participating in the robotics creative contest, taking place the 29th from 10.00-13.00.

The full event agenda is available at:

https://hubs.la/Q04dGkC90

Why is it difficult to reproduce results in robotics?

Even when algorithms are shared, outcomes depend on details that are rarely identical across setups: hardware variations, calibration procedures, environment conditions, and timing, to name a few.

Robots are complex systems, and small differences propagate, leading to diverging behavior.

Standardized platforms help reduce this variance. When experiments are run on comparable hardware and software environments, results become easier to interpret and build upon. This is one of the reasons Duckietown is used in both educational and research settings.

Learn more about Duckietown here: https://hubs.la/Q04bVDMr0

Did you know that Duckietown is being used by researchers in universities and companies in 79 countries to expand our understanding of embodied AI?

We have collected peer-reviewed conference and journal papers selected among 600+ Google Scholar results: https://hubs.la/Q048__nQ0

Are you looking for project ideas for your robotics class?

Review the student projects we have collected from universities worldwide, ranging from autonomous parking implementation to visual language models: https://hubs.la/Q047Tz3H0

Targeted hardware upgrades, broad quality of life improvements!

We have released a chassis upgrade kit that improves driving performance, reduces assembly time and increases compatibility with a range of Jetson Nano kits.

👉 Learn more: https://hubs.la/Q046T7Jc0

Are you getting your robotics class ready for next semester?



Duckietown gives learners hands on experience with robot autonomy and AI on real, programmable vehicles, from first assembly to advanced behaviors.

are now also available pre-assembled and pre-initialized!



Find out how Duckietown can help you teach robot autonomy: https://hubs.la/Q046lrLl0

🎓 Did you know that Open Education Week has just wrapped up?

Remote and free access to educational resources improves accessibility across geographies and demographics.

One way this idea takes shape is through massive open online courses (MOOCs).

To promote hands-on access to AI robotics, in 2020 we launched “Self-Driving Cars with Duckietown”, the first robot autonomy MOOC with hardware, allowing learners to explore autonomous driving through simulation and real robots from anywhere in the world.

Whether you are teaching, learning, or exploring robotics, the course remains freely available on the platform:

https://hubs.la/Q0460D7H0

How do you turn a robot’s design into action? 🤔

The 𝐥𝐨𝐠𝐢𝐜𝐚𝐥 and 𝐩𝐡𝐲𝐬𝐢𝐜𝐚𝐥 architectures must be considered separately:

• 𝐋𝐨𝐠𝐢𝐜𝐚𝐥 architectures define what the robot does, e.g., how perception, planning, and control work.

• 𝐏𝐡𝐲𝐬𝐢𝐜𝐚𝐥 architectures determine how and where the logical architecture code runs, i.e., which CPUs, GPUs, and what are the network links.


During 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭, the logical architecture is mapped to the physical one while balancing latency, bandwidth, and reliability.

Learn about robot autonomy at https://hubs.la/Q044CdvH0.