Abstract
In this workshop, we will learn how to build a simulator for a robot, using the Bevy framework and a physics engine.
To keep the project doable in a day, the robot will be a simple line follower. However, the approach is generalizable to any robot, and the simulation will be extremely accurate, running deterministically in its own time reference and with a high-frequency simulation loop (like 10kHz).
Moreover, the robot configuration and control logic will be implemented in Rust but compiled to WASM, distinctly from the simulator. This clear separation between the simulator and the robot will allow fast iteration in robot experiments. It will also make it easy to collect telemetry data sets related to simulation runs, which in turn will provide deep insight into the robot’s behavior, enabling us to improve it in every iteration.
After this workshop, the participants will know how to:
- Implement a robotic simulator using Bevy.
- Properly virtualize time in the simulation, making it deterministic and as accurate as needed.
- Collect telemetry data sets from the simulation and explore them side by side with a visual rendering of the robot activity.
- Separate and sandbox the robot logic implementation using WASM components.
We will also see how this approach differs from a ROS-based implementation. While it is not possible to also have a hands-on experience of ROS in the same workshop, the tradeoffs between the two approaches will be clearly apparent.
Prerequisites
This workshop will be at the same time playful and deep.
About Rust, it goes in dept on no_std programming and async code execution in embedded context.
Despite the fact that we will not use physical robots, the environment will faithfully simulate the experience of programming a robot and having limited visibility of what is actually happening during its execution.
We will use a data collection (telemetry) system, and see that it is the only reasonable way to gain insight on the causes of any robot “misbehavior” (and, thanks to the realistic physical simulation, there will be plenty of them, which will be fun!).
I have extensive experience in controlling small robots and can guide the participants through all the pitfalls that they’ll find when programming one.
As a side note, I have considered doing the workshop with actual, physical robots, like we did in the 2023 RustLab edition.
And BTW, if you want to repropose that workshop, with those robots, count me in! :-)
However, in this workshop the goal is to teach something more advanced, with faster and more performant robots, where trivial control techniques stop working and smarter ones must be used.
The cost of providing a high-performance line follower to each participant (or even just half of them) would be too high, and I do not think anybody would be interested in sponsoring it.
I can bring with me a real line follower robot, and at the end of the workshop we can see it running and play with it for a while. It’s just not feasible to use it for the actual workshop.
