Choosing an open-source robot arm for AI: SO-ARM100, OpenArm or MeArm?
Three very different open arms, three very different jobs. A quick guide to picking the right one for teaching, data collection or physical-AI research.
Read articleEvery robot we work on starts life in simulation. It is the single biggest lever we have on cost and schedule.
1. Assess
We agree on the task, environment, budget and what "working" means, written down as measurable success criteria.
2. Simulate
We model the robot (usually from the platform's open URDF or CAD) and train or tune its controller in a physics simulator:
- Isaac Lab for large-scale reinforcement learning, as in the Berkeley Humanoid framework
- MuJoCo and Genesis for fast contact-rich manipulation
- Gazebo with ROS 2 for full-system and navigation tests, for example with CHAMP or PX4 SITL
3. Build
We assemble and calibrate the hardware against the simulated baseline. When reality and simulation disagree, we fix the model, not just the robot.
4. Deploy
Field testing, then hand-over: code, documentation and training so your team owns the result.
Want to see this applied to your project? Start a conversation.