RoboDSL

A Domain-Specific Language for GPU-accelerated robotics applications with ROS2 and CUDA

GitHub

Overview

RoboDSL is a Domain-Specific Language (DSL) that simplifies the development of GPU-accelerated robotics applications. It provides a high-level abstraction that automatically generates optimized C++/CUDA code with ROS2 integration, allowing developers to focus on algorithm design rather than low-level boilerplate.

The core idea is straightforward: you write declarative .robodsl files that describe your robotics pipelines—CUDA kernels, ONNX models, ROS2 nodes, and multi-stage processing workflows—and RoboDSL generates the production-ready C++/CUDA infrastructure for you.

Key Features

  • GPU Acceleration: Native CUDA kernel generation with type-safe wrappers and memory management
  • ROS2 Integration: Automatic ROS2 node generation with QoS configuration, lifecycle management, and topic pub/sub
  • ML Model Support: ONNX Runtime integration with optional TensorRT optimization for inference
  • Pipeline Orchestration: Multi-stage processing pipelines that chain preprocessing, CUDA kernels, and ML models
  • Template-Driven Generation: Jinja2-based code generation for maintainability and extensibility

Architecture & Implementation

RoboDSL is built around a modular, AST-first architecture:

  • Lark Parser: Robust grammar parsing with the .robodsl syntax, producing an Abstract Syntax Tree
  • Semantic Analysis: Validation and type-checking before code generation
  • Code Generators: Template-based generation for C++, CUDA, CMake, Python nodes, and ROS2 launch files
  • CLI Tooling: robodsl command-line interface for building projects from .robodsl specs

The project also includes a VS Code extension for syntax highlighting and language support, plus comprehensive tests and documentation.

Example

A minimal RoboDSL pipeline might define a CUDA preprocessing kernel, an ONNX detection model, and a ROS2 node—all in a single declarative file. Running colcon build (or the equivalent RoboDSL build command) generates the full C++/CUDA package, CMake configuration, and launch files.

Tech Stack

  • Languages: Python 3.8+, C++, CUDA
  • Parsing: Lark
  • Code Generation: Jinja2
  • Robotics: ROS2
  • ML / Vision: ONNX Runtime, OpenCV