Why Kompass?

Robotic navigation isn’t about perfecting a single component; it is about architecting a system that survives contact with the real world.

While metric navigation has matured, deploying robots extensively in dynamic environments remains an unsolved challenge. As highlighted by the ICRA BARN Challenges, static pipelines fail when faced with the unpredictability of the physical world:

“A single stand-alone approach that is able to address all variety of obstacle configurations all together is still out of our reach.” — Lessons from The 3rd BARN Challenge (ICRA 2024)

Kompass was built to fill this gap. Unlike existing solutions that rely on rigid behavior trees, Kompass is an event-driven, GPU-native stack designed for maximum adaptability and hardware efficiency.

1. True Adaptive Navigation (Event-Driven Core)

Kompass adheres to open event-driven software standards, allowing the navigation stack to reconfigure itself on the fly.

  • Dynamic Response: The stack adapts not just to internal robot states (like Battery Low), but to external world events (e.g., “Crowd Detected” or “Entering Warehouse”).

  • Context-Aware Control: You can configure the system to use a Pure Pursuit controller on open roads, switch to DWA indoors, and fallback to a precise docking controller, all triggered by environmental context.

  • Simplified Logic: Unlike complex Behavior Trees that can become unmanageable, Kompass allows you to define clean, event-based transitions and fallback behaviors for every component.

2. High-Performance & Vendor-Agnostic GPU Acceleration

Kompass is engineered for speed, utilizing C++ for core algorithms and multi-threading for CPU tasks. However, its standout feature is its approach to hardware acceleration.

  • SYCL-Based Architecture: Kompass is the first navigation framework to support cross-GPU acceleration.

  • Hardware Freedom: Unlike CUDA-locked solutions, Kompass runs natively on Nvidia, AMD, Intel and other integrated GPUs/APUs without device-specific hacks.

  • Parallel Power: Compute-heavy tasks like path planning, control, and map updates are offloaded to the GPU, freeing up your CPU for high-level logic.

3. ML Models as First-Class Citizens

Because Kompass is event-driven, it can directly utilize the outputs of Machine Learning models to drive behavior.

  • Neuro-Symbolic Control: Use an object detection model to trigger a controller switch (e.g., switching to Human-Aware TEB when people are detected).

  • VLM Integration: Use Vision Language Models to answer abstract queries (e.g., “Am I indoors or outdoors?”) and fundamentally alter the robot’s planning strategy based on the answer.

  • Deep Integration: See our tutorial on using Vision Tracking with EmbodiedAgents to see this in action.

4. Pythonic Simplicity, ROS Compatible

We bridge the gap between high-performance C++ and developer-friendly Python.

  • One-Script Configuration: Using Sugarcoat🍬, you can configure a sophisticated, multi-fallback navigation system in a single, readable Python script.

  • Clean Architecture: Core algorithms (Kompass Core) are decoupled from ROS wrappers. This ensures that upgrading ROS distributions (or switching middleware) doesn’t break your underlying navigation logic.

  • Extensible: The modular design allows the community to easily plug in new planners or controllers in Python for prototyping or C++ for production.