4 releases (2 breaking)
0.4.0 | Oct 29, 2024 |
---|---|
0.3.1 | Oct 12, 2024 |
0.3.0 | Sep 30, 2024 |
0.2.3 | Sep 11, 2024 |
#82 in Robotics
368 downloads per month
47KB
841 lines
Copper
Copper is a user-friendly robotics framework designed for creating fast and reliable robots. Copper is to robots what a game engine is to games.
-
Easy: Copper offers a high-level configuration system and a natural Rust-first API.
-
Fast: Copper uses Rust's zero-cost abstractions and a data-oriented approach to achieve sub-microsecond latency on commodity hardware, avoiding heap allocation during execution.
-
Reliable: Copper leverages Rust's ownership, type system, and concurrency model to minimize bugs and ensure thread safety.
-
Product Oriented: Copper aims to avoid late-stage infra integration issues by generating a very predictable runtime.
[!NOTE] Copper is still in early development / alpha stage, and the APIs are subject to change. We are looking for contributors to help us build the best robotics framework possible. If you are interested, please join us on Gitter or open an issue.
Copper has been tested on: Linux (x86_64, armv7, aarch64 & riskv64) and MacOS (arm64).
Technical Overview
Copper is a data-oriented runtime with these key components:
-
Task Graph: Described in RON, this configures the system's topology, specifying which tasks communicate and setting types for nodes and messages.
-
Runtime Generator: This component decides on an execution plan based on the graph's metadata. It preallocates a " Copper List" to maximize sequential memory access during execution.
-
Zero-Copy Data Logging: Records all messages between tasks without copying data, ensuring efficient logging.
-
Fast Structured Logging: Interns and indexes logging strings at compile time, avoiding runtime string construction and ensuring high-speed textual logging.
What features are already implemented?
- Basic task lifecycle interface: Should be relatively stable for you to start contributing new algorithms, sensors, and actuators.
- Runtime generation: Works but is very simple; this is just a BFS type of execution.
- Log reader & structured log reader: Can export data, currently in Rust debug format.
- Components: Those are also good examples if you want to write your own!
Category | Type | Description | Crate Name | |
---|---|---|---|---|
Sensors | Lidar | Velodyne/Ouster VLP16 | cu-vlp16 | |
IMU | WitMotion WT901 | cu-wt901 | ||
ADC/Position | ADS 7883 3MPSPS SPI ADC | cu-ads7883 | ||
Encoder | Generic Directional Wheel encoder | cu-rp-encoder | ||
Actuators | GPIO | Raspberry Pi | cu-rp-gpio | |
Servo | Lewansoul Servo Bus (LX-16A, etc.) | cu-lewansoul | ||
DC Motor Driver | Half-H Driver for CD Motors | cu-rp-sn754410 | ||
Monitors | TUI Monitor | Console based monitor | cu-consolemon | |
Algorithms | PID Controller | PID Controller | cu-pid |
What features are missing? What do we plan to implement next?
Here are some of the features we plan to implement next (in ~order of priority), if you are interested in contributing on any of those, please let us know!:
To Reach Beta:
- Merging: Add a feature to merge messages from multiple sources as an input to a task.
- Monitoring: We need a parallel system that can listen to monitoring messages and act accordingly.
- Batching/Aligning: add a feature to batch messages for high frequency sources to reduce the number of Copper Lists.
- Deterministic log replay + Sim: As the runtime is generated in a deterministic fashion, we need to add hooks to inject messages into an existing runtime.
- Parallel Copper Lists: Today Copper is monothreaded; this should enable concurrent Copper Lists to be executed at the same time with no contention.
- ROS/DDS interfacing: Build a pair of sink and source to connect to existing ROS systems, helping users migrate their infra bit by bit.
To Reach RC1:
- Extensible scheduling: Enables a way to give hints to copper to schedule the workload
- Modular Configuration: As robots built with Copper gain complexity, users will need to build "variations" of their robots without duplicating their entire RON file.
- Distributed Copper: Currently, we can only create one process. We need proper RPC filtering copper lists per subsystem. So we are only at the beginning, but so much cool stuff is coming up!
Kickstarting a copper project for the impatients
You can generate a project from a template present in the repo. It will ask you the name you want to pick interactively.
cargo install cargo-generate
git clone https://github.com/copper-project/copper-rs
cd copper-rs/templates
cargo cunew [path_where_you_want_your_project_created]
🤷 Project Name:
Check out copper-templates for more info.
How does a Copper application look like?
Here is a simple example of a task graph in RON:
(
tasks: [
(
id: "src", // this is a friendly name
type: "FlippingSource", // This is a Rust struct name for this task see main below
),
(
id: "gpio", // another task, another name
type: "cu_rp_gpio::RPGpio", // This is the Rust struct name from another crate
config: { // You can attach config elements to your task
"pin": 4,
},
),
],
cnx: [
// Here we simply connect the tasks telling to the framework what type of messages we want to use.
(src: "src", dst: "gpio", msg: "cu_rp_gpio::RPGpioMsg"),
],
Then, on your main.rs:
// Your application will be a struct that will hold the runtime, loggers etc.
// This proc macro is where all the runtime generation happens. You can see the code generated by the macro at
// compile time.
#[copper_runtime(config = "copperconfig.ron")] // this is the ron config we just created.
struct MyApplication {}
// Here we define our own Copper Task
// It will be a source flipping a boolean
pub struct FlippingSource {
state: bool,
}
// You need to provide at least "new". But you have other hooks in to the Lifecycle you can leverage
// to maximize your opportunity to not use resources outside of the critical execution path: for example start, stop,
// pre_process, post_process etc...
impl CuTaskLifecycle for FlippingSource {
fn new(_config: Option<&copper::config::ComponentConfig>) -> CuResult<Self>
where
Self: Sized,
{
Ok(Self { state: true })
}
}
// We implement the CuSrcTask trait for our task as it is a source / driver (with no internal input from Copper itself).
impl<'cl> CuSrcTask<'cl> for FlippingSource {
type Output = output_msg!('cl, RPGpioPayload);
// Process is called by the runtime at each cycle. It will give:
// 1. the reference to a monotonic clock
// 2. a mutable reference to the output message (so no need to allocate of copy anything)
// 3. a CuResult to handle errors
fn process(&mut self, clock: &RobotClock, output: Self::Output) -> CuResult<()> {
self.state = !self.state; // Flip our internal state and send the message in our output.
output.payload = RPGpioPayload {
on: self.state,
creation: Some(clock.now()).into(),
actuation: Some(clock.now()).into(),
};
Ok(())
}
}
fn main() {
// Copper uses a special log format called "unified logger" that is optimized for writing. It stores the messages between tasks
// but also the structured logs and telemetry.
// A log reader can be generated at the same time as the application to convert this format for post processing.
let logger_path = "/tmp/mylogfile.copper";
// This basic setup is a shortcut to get you running. If needed you can check out the content of it and customize it.
let copper_ctx =
basic_copper_setup(&PathBuf::from(logger_path), true).expect("Failed to setup logger.");
// This is the struct logging implementation tailored for Copper.
// It will store the string away from the application in an index format at compile time.
// and will store the parameter as an actual field.
// You can even name those: debug!("This string will not be constructed at runtime at all: my_parameter: {} <- but this will be logged as 1 byte.", my_parameter = 42);
debug!("Logger created at {}.", logger_path);
// A high precision monotonic clock is provided. It can be mocked for testing.
// Cloning the clock is cheap and gives you the exact same clock.
let clock = copper_ctx.clock;
debug!("Creating application... ");
let mut application =
MyApplication::new(clock.clone(), copper_ctx.unified_logger.clone())
.expect("Failed to create runtime.");
debug!("Running... starting clock: {}.", clock.now()); // The clock will be displayed with units etc.
application.run().expect("Failed to run application.");
debug!("End of program: {}.", clock.now());
}
But this is a very minimal example for a task, please see lifecycle for a more complete explanation of a task lifecycle.
Deployment of the application
Check out the deployment page for more information.
How is it better or different from ROS?
Performance
In the example directory, we have 2 equivalent applications. One written in C++ for ROS and a port in Rust with Copper.
examples/cu_caterpillar
examples/ros_caterpillar
You can them out either just logging on a desktop or with GPIOs on a RPi and you should see a couple order of magnitude difference in performance.
Copper has been design for performance first. Unlike a game engine we use a data oriented approach to minimize latency and maximize throughput.
Safety
As Copper is written in Rust, it is memory safe and thread safe by design. It is also designed to be easy to use and to avoid common pitfalls.
As we progress on this project we plan on implementing more and more early warning to help you avoid "the death by a thousand cuts" that can happen in a complex system.
Dependencies
~5–15MB
~196K SLoC