An open-source, full-stack platform (hardware + software) designed from the ground up for high-performance, secure, and modular robots.
The future of robotics—agile humanoids, adaptive industrial arms, and autonomous field robots—is bottlenecked by its foundational software. Tools like ROS, while revolutionary for research, were not designed for the demands of the modern physical world:
- Poor Real-Time Performance: Millisecond-level jitter and a lack of deterministic guarantees make high-speed, dynamic control (like 1000Hz motion) nearly impossible.
- No "Plug & Play" Modularity: Adding or swapping a limb on a live robot without rebooting the entire system is a nightmare. This stifles innovation and adaptability.
- Inherent Insecurity: A flat, open communication model means any node can potentially disrupt critical control loops, posing a significant safety risk.
- System Complexity: Developers are forced to become experts in dozens of low-level protocols (CAN, EtherCAT, SPI) and manage a fragile web of drivers and dependencies.
We believe robotics needs a new foundation. Not another patch on an old system, but a complete rethinking from first principles.
FluxTendril is that new foundation. It is a vertically-integrated, open-source platform designed to be the central nervous system for next-generation robots. We achieve this through a unique, soft-hard co-design approach:
- ⚡️ Extreme Real-Time: A dual-domain architecture with a deterministic core running at 500-1000Hz for motion control, completely isolated from non-real-time tasks.
- 🔌 True Hot-Swapping: "Actuator/Sensor Bridges" are smart hardware nodes that allow for the safe, live "plug and play" of limbs and sensors, just like a USB device.
- 🔒 Security by Design: From hardware root-of-trust to capability-based APIs, security is a foundational layer, not an afterthought.
- Unified Architecture: We abstract away the chaos of low-level protocols, providing developers with a clean, unified, high-level API.
This project is the spiritual successor to our earlier work, FluxWeave, which validated the community's need for a better real-time communication framework.
(This is a placeholder for the system architecture diagram)
This is the core hardware innovation. Each Bridge is a small, powerful MPU/MCU-based board that sits between the central controller and a set of actuators/sensors.
- Protocol Unification: Converts CAN, EtherCAT, SPI, PWM, etc., into a single, time-stamped data stream.
- Edge Pre-processing: Handles local tasks like sensor filtering, state estimation, and running motor-specific drivers, offloading the central CPU.
- Hot-Swap Management: Contains a partial URDF model and self-identifies on the network, allowing the central controller to rebuild the robot's model in real-time.
- Hardware Security: Features a Secure Element (SE) or TPM for identity and secure boot, preventing unauthorized modules from joining the network.
The central controller's software is split into two isolated domains to guarantee performance and safety.
- Real-Time Core (RTCore):
- Deterministic 500-1000Hz control loop.
- Lock-free data structures for zero-latency data exchange.
- Hardened, minimal kernel (e.g., RT-Linux, Zephyr, or a hypervisor-based approach).
- Non-Real-Time Domain (NRT-Domain):
- Runs high-level processes like AI models (VLM/VLN), path planning, and cloud communication.
- Communicates with the RTCore via a strictly-controlled, asynchronous API.
We use TSN as the communication backbone, providing features unavailable in standard Ethernet or DDS.
- Nanosecond-level Clock Sync (802.1AS): All modules share a unified, hardware-synchronized sense of time.
- Guaranteed Bandwidth & Latency (802.1Qbv): Critical control data is scheduled into protected time slots, immune to network congestion from cameras or LiDAR.
- Enhanced Security (802.1Qci/MACsec): Provides hardware-level filtering and encryption of network traffic.
FluxTendril is currently in the early architectural and prototyping phase. Our roadmap is ambitious and community-driven.
-
[Phase 1] Core Architecture & Prototyping (Current):
- Finalize system architecture and technology choices.
- Develop initial schematics for a low-cost MCU-based Bridge.
- Implement a basic RTCore on a reference hardware platform (e.g., Raspberry Pi CM4 with RT_PREEMPT patch).
- Establish a basic TSN network for clock synchronization.
-
[Phase 2] Minimum Viable Product (MVP):
- A central controller and two Bridges demonstrating real-time synchronized motion.
- A successful hot-swap of one Bridge without interrupting the other's operation.
- Release of the initial Python API for basic control.
-
[Phase 3] Alpha Release & Community Building:
- Support for more complex sensors (e.g., IMUs, Force/Torque sensors).
- Release documentation and tutorials for early adopters.
- Develop a high-performance MPU-based Bridge for vision sensors.
We are actively seeking passionate developers, roboticists, and hardware engineers to help build the future of robotics. Whether you are an expert in real-time kernels, a wizard with FPGAs, or a master of mechanical design, your contribution is valuable.
- Join the Discussion: Check out our Issues and Discussions tabs to see what we're working on and share your ideas.
- Read the Docs: (Coming soon) Dive into our technical documentation to understand the architecture in depth.
- Pick an Issue: Look for issues tagged
good first issueto get started. - Spread the Word: Star the repository and share it with your network!
This project is licensed under the Apache License 2.0. We chose a permissive license to encourage maximum adoption and commercial use.
Let's build the nervous system that will finally let robots run.