I’ve been building Eigen Robotics — a Python-first, ML-native robotics framework designed to unify simulation and real-world control under one clean, config-driven API.
Think of it as PyTorch meets robotics infra — focused on simplicity, composability, and machine learning integration from day one.
Modern robotics development is fragmented:
Every simulator and robot driver lives in its own ecosystem.
“Train in Sim A → validate in Sim B → deploy on hardware” usually means rewriting your code.
Integrating ML pipelines (RL, IL, vision models) is brittle and non-standard.
ROS is powerful, but often overkill for lightweight experimentation or ML research.
What Eigen Does
Eigen focuses on lightweight modularity and YAML-based configuration, so you can define robots, sensors, and simulators declaratively — no boilerplate.
Unified config + runtime system using OmegaConf
Cross-simulator API (PyBullet, MuJoCo, Genesis)
Real-world bridge layer tested with Franka and Unitree
Lightweight LCM-based communication (ROS-like, but faster and simpler)
CLI and Python APIs for launching and managing nodes
Built-in RL and data collection loops
Everything can be imported and used directly from Python (import eigen), or launched via CLI (eigen launch viper.yaml).
We’re preparing our next release focused on:
Better sim-to-real transfer tools
Plugin system for new robots/sensors
API and documentation polish
We’re looking for researchers, engineers, and roboticists interested in trying Eigen and giving feedback on real-world workflows — especially around ML integration and multi-simulator setups.
If this sounds relevant to your work, I’d love to chat or share early access. Feedback is super welcome — especially around pain points with ROS, sim-real transfer, or existing infra.
I’ve been building Eigen Robotics — a Python-first, ML-native robotics framework designed to unify simulation and real-world control under one clean, config-driven API. Think of it as PyTorch meets robotics infra — focused on simplicity, composability, and machine learning integration from day one.
Modern robotics development is fragmented: Every simulator and robot driver lives in its own ecosystem. “Train in Sim A → validate in Sim B → deploy on hardware” usually means rewriting your code. Integrating ML pipelines (RL, IL, vision models) is brittle and non-standard. ROS is powerful, but often overkill for lightweight experimentation or ML research. What Eigen Does Eigen focuses on lightweight modularity and YAML-based configuration, so you can define robots, sensors, and simulators declaratively — no boilerplate. Unified config + runtime system using OmegaConf Cross-simulator API (PyBullet, MuJoCo, Genesis) Real-world bridge layer tested with Franka and Unitree Lightweight LCM-based communication (ROS-like, but faster and simpler) CLI and Python APIs for launching and managing nodes Built-in RL and data collection loops Everything can be imported and used directly from Python (import eigen), or launched via CLI (eigen launch viper.yaml).
We’re preparing our next release focused on: Better sim-to-real transfer tools Plugin system for new robots/sensors API and documentation polish
We’re looking for researchers, engineers, and roboticists interested in trying Eigen and giving feedback on real-world workflows — especially around ML integration and multi-simulator setups. If this sounds relevant to your work, I’d love to chat or share early access. Feedback is super welcome — especially around pain points with ROS, sim-real transfer, or existing infra.